zrj [Tue, 11 Oct 2016 07:28:51 +0000 (10:28 +0300)]
ncurses: Trim sources that create empty objects.
zrj [Tue, 11 Oct 2016 07:27:24 +0000 (10:27 +0300)]
libssh: Trim sources that create empty objects.
zrj [Tue, 11 Oct 2016 07:16:31 +0000 (10:16 +0300)]
libpam: Trim sources that create empty objects.
zrj [Tue, 11 Oct 2016 07:13:12 +0000 (10:13 +0300)]
libmagic: Trim sources that create empty objects.
zrj [Tue, 11 Oct 2016 06:22:10 +0000 (09:22 +0300)]
libedit: Trim sources that create empty objects.
zrj [Wed, 5 Oct 2016 11:32:04 +0000 (14:32 +0300)]
gcc50: Trim sources that create empty objects.
* Disable gen source of HAVE_peephole, only applies for gcc <= 5.x.
* Exclude {hw,loop}-doloop.c, protected by HAVE_doloop_end guard (gcc < 6).
zrj [Wed, 5 Oct 2016 08:21:42 +0000 (11:21 +0300)]
libmgmp: Trim sources that create empty objects.
zrj [Wed, 5 Oct 2016 08:21:00 +0000 (11:21 +0300)]
libmpfr: Trim sources that create empty objects.
zrj [Wed, 5 Oct 2016 07:57:11 +0000 (10:57 +0300)]
grep(1): Trim sources that create empty objects.
There are several things that need to be addressed here.
zrj [Fri, 21 Oct 2016 06:05:26 +0000 (09:05 +0300)]
sys/boot: Remove empty source.
zrj [Fri, 21 Oct 2016 10:35:12 +0000 (13:35 +0300)]
bootstrap: Add workaround for chflagsat() from 4.0.
Avoid failing syscalls during installworld while updating from DragonFly 4.0.
Add special private version of chflagsat() to aid while performing migration.
Add triple protection for this private implementation variant.
Hopefully chflags-compat.h would never need any newer additions.
Since there are other issues that require both installkernel and installworld,
standard installkernel, reboot and installworld is not possible while updating
to recent DragonFly releases. chflags(1) must be handled with care.
Tested-on: kern.osreldate == 400000 updating straight to master (4.7).
installworld step fail only on single non fatal chflags: invalid flag: xlink
For make upgrade to work sendmail must be removed from /etc/mail/mailer.conf
reboot(8) must be used cause shutdown is only functional only on new kernel.
Submitted-by: dillon@
zrj [Fri, 21 Oct 2016 10:20:13 +0000 (13:20 +0300)]
bootstrap: Fix old boostrapping issue once and for all part 2.
In
f08f1baa605ef8f4a93a42e1d790a9fcfb44df35 I forgot to git add -p the line to
export WORLD_VERSION variable to make it constant through buildworld process.
This time add it close so it would not get lost again, fail window was 1 month.
make buildworld WORLD_VERSION=400300 still have worked.
While there, add bootstrap workaround to bootstrap from 4.0-RELEASE too.
chflags(1) issue with chflagsat() is a bit more subtle, will be pushed later.
Tested-on: dfly-x86_64-4.0.6_REL.img updating straight to 4.7-DEVELOPMENT.
Not bad considering DragonFly 4.0 had gcc44 + gcc47 and was still able to
bootstrap the DragonFly 4.7 having gcc v5.0 as main compiler.
Reported-by: dillon@
zrj [Thu, 20 Oct 2016 09:50:58 +0000 (12:50 +0300)]
<wchar.h>: Add support for compilers -fshort-wchar.
Both clang and gcc need to be able to override wchar_t type.
zrj [Fri, 14 Oct 2016 13:41:31 +0000 (16:41 +0300)]
installer: Init payload to NULL.
Code checks for payload if it is allocated or not, so technically
this is a case where this pointer is used not initialized.
zrj [Wed, 12 Oct 2016 05:19:49 +0000 (08:19 +0300)]
ppp: Avoid gcc warnings.
ncprange_getip4mask() will not initialize mask for cases other than AF_INET
so give default value of INADDR_ANY (0x0) even if compiler misses that
ncprange_getip4mask() called only for AF_INET case in main function.
zrj [Tue, 11 Oct 2016 14:55:47 +0000 (17:55 +0300)]
setkey(8): Fix m_len type to match real one.
Given that it is used in places for sizeof() etc it should be unsigned.
zrj [Tue, 11 Oct 2016 14:47:17 +0000 (17:47 +0300)]
mlxcontrol(8): Provide defaults to few variables.
GCC is right here, mlxd_find_ctrlr() will not initiallize these on failure.
zrj [Tue, 11 Oct 2016 14:26:19 +0000 (17:26 +0300)]
telnet(1): Fix line definition to match extern one.
In telnetd it is extern char[16] so just use that.
zrj [Tue, 11 Oct 2016 14:19:39 +0000 (17:19 +0300)]
wall(1): Avoid gcc warning.
zrj [Tue, 11 Oct 2016 14:16:11 +0000 (17:16 +0300)]
users(1): Avoid gcc warning.
zrj [Tue, 11 Oct 2016 14:07:08 +0000 (17:07 +0300)]
systat(1): Fix curscale type to match other one.
Also avoid triggering the the warning for ep.
zrj [Tue, 11 Oct 2016 14:00:58 +0000 (17:00 +0300)]
usr.bin/rfcomm_sppd: Avoid gcc warnings.
zrj [Tue, 11 Oct 2016 13:25:33 +0000 (16:25 +0300)]
cal(1): Avoid gcc warnings.
Looks like here the compiler is right.
zrj [Tue, 11 Oct 2016 09:44:26 +0000 (12:44 +0300)]
sbin/quotacheck: Rename maxino to remove globals conflict.
fsck/fsck.h has it as ufs1_ino_t so to avoid type mismatches rename it.
zrj [Mon, 10 Oct 2016 10:00:29 +0000 (13:00 +0300)]
adventure(6): Fix type to match public one.
zrj [Mon, 10 Oct 2016 09:59:58 +0000 (12:59 +0300)]
dm(8): Avoid warning on gcc with -flto.
zrj [Mon, 10 Oct 2016 07:52:29 +0000 (10:52 +0300)]
talkd(8): Avoid warning on gcc with -flto.
zrj [Tue, 11 Oct 2016 13:38:42 +0000 (16:38 +0300)]
less(1): Fix type mismatches in different objects.
Not sure which one is better int or long. For now downgrade to int,
cause it is less changes in contrib code.
zrj [Tue, 11 Oct 2016 11:13:51 +0000 (14:13 +0300)]
cvs: Use mempcpy() from libc.
Helps with LTO builds.
zrj [Tue, 11 Oct 2016 09:20:51 +0000 (12:20 +0300)]
devd(8): Make it possible to compile with LTO.
Make -Wodr warning (internal to c++ headers) under -flto non fatal.
Also give initial value, pidfile_open() has too many return points.
zrj [Mon, 10 Oct 2016 09:43:25 +0000 (12:43 +0300)]
backgammon(6): Prevent LTO seeing details for now.
init.c should contain "back.h" for types (TBI later).
-flto gives lots of error: type of 'args' does not match original declaration [-Werror]
/usr/src/games/backgammon/backgammon/../common_source/back.h
extern char *args[16]; /* args passed to teachgammon and back */
/usr/src/games/backgammon/backgammon/../common_source/init.c
char args[100];
/usr/src/games/backgammon/backgammon/../common_source/back.h
extern int old; /* original tty status */
/usr/src/games/backgammon/backgammon/../common_source/init.c
struct termios tty, old, noech, raw;
/usr/src/games/backgammon/backgammon/../common_source/back.h
extern int raw; /* raw tty status, no echo */
/usr/src/games/backgammon/backgammon/../common_source/init.c
struct termios tty, old, noech, raw;
/usr/src/games/backgammon/backgammon/../common_source/back.h
extern int noech; /* original tty status without echo */
/usr/src/games/backgammon/backgammon/../common_source/init.c
struct termios tty, old, noech, raw;
lto1: all warnings being treated as errors
zrj [Wed, 5 Oct 2016 11:35:30 +0000 (14:35 +0300)]
csh(1): Allow to bootstrap with -flto.
tcsh(1) replaces malloc, thus with LTO sees multiple symbols.
zrj [Tue, 4 Oct 2016 05:16:34 +0000 (08:16 +0300)]
install(1): Silence false gcc warnings.
Happens on LTO builds.
Imre Vadász [Sun, 4 Sep 2016 08:51:11 +0000 (10:51 +0200)]
if_iwm - Process multiple frames per RX buffer.
* Factor out iwm_handle_rxb() function from iwm_notif_intr().
* Removing the IWM_FH_RCSR_CHNL0_RX_CONFIG_SINGLE_FRAME_MSK flag allows
the device to put multiple frames (both command responses and 80211
frames) into a single RX buffer.
* Uses m_copym() to split up the receive buffers when multiple 80211
frames are received in one RX buffer. The effect is basically the same
as when using m_split(), but we want to keep the original mbuf around
when calling iwm_mvm_rx_rx_mpdu() to make error handling a bit easier
for now.
* Contains a small optimization to avoid the m_copym() when only a single
80211 frame is received in one RX buffer (i.e. matching the existing
behaviour).
Taken-From and Inspired-By: Linux iwlwifi
Sascha Wildner [Tue, 18 Oct 2016 21:13:17 +0000 (23:13 +0200)]
Remove dead /usr/include/compat symlink, too.
Followup commit to
6737a4ffb92.
Sascha Wildner [Tue, 18 Oct 2016 20:50:13 +0000 (22:50 +0200)]
kernel: Remove the COMPAT_43 kernel option along with all related code.
It is commented out in our default kernel config files for almost five
years now, since
9466f37df5258f3bc3d99ae43627a71c1c085e7d.
Approved-by: dillon
Dragonfly-bug: <https://bugs.dragonflybsd.org/issues/2946>
Sascha Wildner [Tue, 18 Oct 2016 19:34:03 +0000 (21:34 +0200)]
bsd-family-tree: Sync with FreeBSD (FreeBSD 11.0).
Sascha Wildner [Tue, 18 Oct 2016 19:32:52 +0000 (21:32 +0200)]
Update the pciconf(8) database.
October 13, 2016 snapshot from http://pciids.sourceforge.net/
Sascha Wildner [Tue, 18 Oct 2016 19:27:32 +0000 (21:27 +0200)]
<sys/types.h>: Add definitions of bool, true and false for the kernel.
Remove some local definitions in drivers and also switch boolean_t
over to _Bool (if available).
This is similar to how FreeBSD does it.
zrj [Tue, 18 Oct 2016 15:37:21 +0000 (18:37 +0300)]
tcpdchk(8): Honor global CFLAGS.
Forgot to include in
8d378610e3b5687c707bc8aad4e11a3a96bea2fc for tcpdmatch(8).
zrj [Tue, 18 Oct 2016 13:07:16 +0000 (16:07 +0300)]
binutils225: Honor global CFLAGS.
Make CFLAGS partially overridable so to preserve global CFLAGS.
block2/ld has its own config.h but still requires bfd.h.
Ensure that private config.h takes precedence over general bulibs one.
While testing noticed that block2/gas ignores its config.h, so fix that too.
block2/libgold is fine, since it uses block3/Makefile.gld.
Sepherosa Ziehau [Tue, 18 Oct 2016 13:23:12 +0000 (21:23 +0800)]
route: addroute may change rt->rt_ifa; keep the local copy in sync.
Reported-by: pecan
Sepherosa Ziehau [Fri, 9 Sep 2016 00:55:48 +0000 (08:55 +0800)]
x86_64: Work around AMD erratum 793 for family 16h, models 00h-0Fh
Obtained-from: FreeBSD r305539
zrj [Tue, 18 Oct 2016 11:51:38 +0000 (14:51 +0300)]
tcpdmatch(8): Honor global CFLAGS.
It does not look like override was on purpose.
zrj [Tue, 18 Oct 2016 11:38:54 +0000 (14:38 +0300)]
restore(8): Remove (void *) casts.
Mainly to unbreak gcc47 buildworld.
Reported-by: swildner
zrj [Wed, 12 Oct 2016 08:24:28 +0000 (11:24 +0300)]
restore(8): Fix extern type mismatch.
Noticed while playing with LTO.
While there, add volatile to prevent clobbering @ WORLD_CCOPTLEVEL=g
zrj [Tue, 18 Oct 2016 09:01:17 +0000 (12:01 +0300)]
restore(8): Fix and enable WARNS.
* Add most cases were const char * should have been used.
* In few cases use __DECONST(), should be fine for these.
* Suffix variables with 'g' and 'l' to prevent shadowing.
* Mark unused parameters.
* Convert swabst() to take const char * instead of u_char *.
No functional change intended, utility still works.
(dump -af dumpas /dev/da1s1 && restore -if dumpas) restores files.
zrj [Tue, 18 Oct 2016 08:35:17 +0000 (11:35 +0300)]
restore(8): Use size_t were appropriate.
Mostly WARNS=3 cleanup.
Also add vlatile to prevent clobbering.
zrj [Tue, 18 Oct 2016 08:06:59 +0000 (11:06 +0300)]
restore(8): Fix formats in prints.
Also, exlude two prints if NAME_MAX >= 255.
Taken-from: FreeBSD
zrj [Tue, 18 Oct 2016 07:24:33 +0000 (10:24 +0300)]
restore(8): General cleanup.
* Fix include order.
* Add missing va_end().
Imre Vadász [Mon, 17 Oct 2016 22:20:29 +0000 (00:20 +0200)]
if_iwm - Change UCODE_TLV_API #define-s from bitmasks to indexes.
* Fixes oversight from commit
757eecf0e6c92745aa2eee95811e573c8300850e.
fw_has_api now uses the isset macro instead of a simple logical-and.
Imre Vadász [Fri, 16 Sep 2016 21:43:51 +0000 (23:43 +0200)]
if_iwm - Sync iwm_read_firmware()'s loop to iwlwifi's code.
Taken-From: Linux iwlwifi
Imre Vadász [Tue, 6 Sep 2016 19:45:07 +0000 (21:45 +0200)]
if_iwm - change the check for ADD_STA status, use IWM_ADD_STA_STATUS_MASK.
Taken-From: Linux iwlwifi (git
837c4da98481d4e504b2aba077c8528fee1b6d13)
Imre Vadász [Mon, 17 Oct 2016 21:17:06 +0000 (23:17 +0200)]
if_iwm - Sanity check channel for IEEE80211_CHAN_ANYC in if_iwm_mac_ctxt.c.
* This avoids panicing in some broken vap state handling cases.
Imre Vadász [Sun, 16 Oct 2016 19:36:15 +0000 (21:36 +0200)]
if_iwm - Get rid of another usage of the IWM_DEFAULT_MACID/_COLOR constant.
Matthew Dillon [Mon, 17 Oct 2016 19:15:57 +0000 (12:15 -0700)]
kernel - Fix panic/crash in vm_object_page_remove()
* When vm_object_page_remove() is called with one page, the caller is allowed
to hold a spinlock (typically on that page). The callback is not supposed
to call user_yield in that situation.
* Fix missing initialization of info.count which could result in the callback
issuing a yield when it was not supposed to.
Reported-by: htse
zrj [Wed, 12 Oct 2016 13:31:32 +0000 (16:31 +0300)]
sys/boot: Annotate LTO issue in pxe.c
This needs to be addressed for proper use of LTO.
zrj [Fri, 30 Sep 2016 11:23:13 +0000 (14:23 +0300)]
binutils225: Make plugin dir overridable at compile time.
This will help to avoid touching mtree BSD.usr.dist outside libexec,
will simplify LTO plugins(liblto_plugin.so, LLVMgold.so)
installing(symlinking) and auto loading in ar, nm, ranlib.
zrj [Mon, 17 Oct 2016 07:44:39 +0000 (10:44 +0300)]
Merge branch 'vendor/BINUTILS225'
zrj [Fri, 30 Sep 2016 10:07:50 +0000 (13:07 +0300)]
Import missing source needed to enable plugin support in utilities.
zrj [Mon, 17 Oct 2016 06:18:11 +0000 (09:18 +0300)]
Remove profiled version of libraries too.
* Expand OpenSSL removal list to include profiled versions.
* Remove libprivate_panel ones after
19ba4072a64db34c2a861d48aa5ffe33155826ef.
John Marino [Sun, 16 Oct 2016 22:34:39 +0000 (17:34 -0500)]
Step 1/2 of OpenSSL remove: Suppress building
As of now, the NO_OPENSSL directive has been replaced with an undocumented
FORCE_OPENSSL directive. Unless FORCE_OPENSSL is defined in make.conf,
the OpenSSL libraries, headers, and man pages will no longer be built and
installed.
Existing OpenSSL files will remain installed unless REMOVE_OPENSSL_FILES
is defined for the upgrade command, e.g.
> make upgrade REMOVE_OPENSSL_FILES=yes
It is recommended that all remaining programs and libraries linked to the
base OpenSSL libraries be rebuilt to use one of the DPorts SSL libraries.
The DPorts packages have been built with DPorts LibreSSL for a few weeks
now, so ensuring all installed packages are reinstalled or rebuilt will be
sufficient for most users, but don't forget about software built outside
of DPorts.
The OpenSSL sources and makefiles will be removed within a month (step 2),
but FORCE_OPENSSL will work in the meantime to continue building OpenSSL
although it will receive no further updates.
Matthew Dillon [Sun, 16 Oct 2016 21:36:10 +0000 (14:36 -0700)]
kernel - Modestly reduce default maxvnodes calculation
* Change the multipliers to modestly reduce the default kern.maxvnodes
calculation. For example, maxvnodes on a machine with 8GB of ram will
be reduced from 598103 to 478483.
* Primarily targetted to hammer filesystems which can wind up using an
unnecessarily huge amount of kernel memory caching vnode-related
structures.
Matthew Dillon [Sun, 16 Oct 2016 21:31:40 +0000 (14:31 -0700)]
kernel - Reduce excessive inode hash table allocations
* Reduce excessive inode hash table allocations in various filesystems
(primarily ufs). Introduce vfs_inodehashsize() to calculate a reasonable
hash table size instead of using 'maxvnodes'.
* The new formula is to generally use maxvnodes / 2 (2 x stacking for the
chained hash table). When maxvnodes is large we use maxvnodes / 4, and
if maxvnodes is very large (> 1M vnodes) we use maxvnodes / 8. This
significantly reduces the amount of kernel memory used when mounting
ufs, ext2fs, hpfs, isofs, msdosfs, nfs, ntfs, and smbfs filesystems.
Matthew Dillon [Sun, 16 Oct 2016 21:23:44 +0000 (14:23 -0700)]
tmpfs - Fix mount_tmpfs -C
* Fix mount_tmpfs -C. FSCopy requires the contents of the pointer
whos address is passed into it to be properly initialized (to NULL
in this case).
* Fixes -O2 compilation of rtld & mount_tmpfs which were not problems
with -O2 itself but instead changed the uninitialized garbage of the
local variable we fixed above.
reported-by: zrj
Matthew Dillon [Sun, 16 Oct 2016 21:13:39 +0000 (14:13 -0700)]
kernel - Rename desiredvnodes to maxvnodes, fix deadlock
* Rename the kernel variable 'desiredvnodes' to 'maxvnodes' to match
the sysctl name (which has always been 'maxvnodes'), and to make the
code more readable.
* Probable fix to a rare mount/umount deadlock which can occur in two
situations (1) When a large number of mounts and unmounts are running
concurrently, and (2) During a umount -a, shutdown, or reboot.
* Considered minor, normal use cases will not reproduce this bug. Only
synth or poudriere can generate the mount/umount traffic necessary to
reproduce this bug.
* Also fixes a minor kernel memory leak of the mount structure which can
occur when a 'df' or filesystem sync races a umount. Also minor.
Reported-by: marino (mount race)
Imre Vadász [Sun, 16 Oct 2016 18:55:13 +0000 (20:55 +0200)]
if_iwm - Partly sync if_iwm_binding.c to Linux iwlwifi code.
* Store macid and color values in struct iwm_vap, to avoid hardcoded
constants a bit.
* Add iwm_mvm_binding_remove_vif() function (will be used in disconnecting
from an access point without resetting the whole device).
* Not adding code from Linux iwlwifi yet, to handle one PHY context to
be bound to several VAPs/virtual-interfaces, it's definitely not needed
in the near future.
Taken-From: Linux iwlwifi
Imre Vadász [Sun, 16 Oct 2016 10:23:44 +0000 (12:23 +0200)]
if_iwm - Move in_phyctxt from struct iwm_node to phy_ctxt in struct iwm_vap.
* This better matches how things are organized in Linux's iwlwifi.
Tomohiro Kusumi [Sun, 16 Oct 2016 04:01:54 +0000 (13:01 +0900)]
sbin/mount_hammer: Add missing switch case from
bab16452
The previous commit
bab16452 needed 1..15 case for expected values,
so as not to show an irrelevant message via default case.
Tomohiro Kusumi [Sun, 16 Oct 2016 03:07:51 +0000 (12:07 +0900)]
sbin/mount_hammer: Fix warning messages
Add warning messages for -1 and non-0-15 for master_id.
-1 and 0-15 inclusive are the only supoprted values,
so remove probably from the existing warning message.
Tomohiro Kusumi [Sat, 15 Oct 2016 14:46:57 +0000 (23:46 +0900)]
sbin/hammer: Mention bug in hammer recover command
newfs_hammer overwrites layer1/2, but doesn't overwrite data in
blockmap beyond layer1/2, so old data is still there if the volume
has been re-formatted until the new fs overwrites old fs by using
more disk space than the old fs.
This is usually not a problem because the new fs doesn't have a
pointer to data in old fs, but hammer recover just iterates zone-2
offset till it hits end of the disk space which may include data
from the old fs. As mentioned in the comment, it's likely to hit
assertion if the old data located near the end of the volume is
B-Tree nodes depending on situation of the old fs.
Tomohiro Kusumi [Sat, 15 Oct 2016 14:16:14 +0000 (23:16 +0900)]
sbin/hammer: Add volume signature check on loading a volume
Tomohiro Kusumi [Sat, 15 Oct 2016 01:15:02 +0000 (10:15 +0900)]
sbin/hammer: Cleanup arguments of init_volume()
Change the order of arguments to make the i/f similar to open(2)
as well as load_volume().
This is only for newfs (and volume-add), nothing else gets affected.
Tomohiro Kusumi [Fri, 14 Oct 2016 12:33:59 +0000 (21:33 +0900)]
sbin/hammer: Make hammer recover use hammer_parsedevs_noverify()
hammer recover command should be able to run the recovery code
even if volume header is broken (not just broken blockmap part)
to certain extent, since the purpose of this is to recover contents
of a broken filesystem.
By using 0 for verify argument of load_volume(), hammer recover
skips sanity checks for ondisk volume header.
Also see
173bd31c.
Tomohiro Kusumi [Fri, 14 Oct 2016 12:27:09 +0000 (21:27 +0900)]
sbin/hammer: Add verify argument to load_volume()
Add an argument to select whether to verify volume header or not.
This is for the next commit.
Matthew Dillon [Fri, 14 Oct 2016 16:32:12 +0000 (09:32 -0700)]
kernel - Fix improper user-space access in sys___semctl()
* Fix an improper user-space access in sys__semctl()
* Fix an improper kernel-space access that was using
a user-supplied pointer.
Reported-by: Mateusz Kocielski - LogicalTrust
François Tigeot [Fri, 14 Oct 2016 06:59:09 +0000 (08:59 +0200)]
drm/linux: Add kmap_to_page()
François Tigeot [Thu, 13 Oct 2016 10:07:00 +0000 (12:07 +0200)]
drm/linux: Add get_nr_swap_pages()
Imre Vadász [Wed, 12 Oct 2016 20:19:31 +0000 (22:19 +0200)]
if_iwm - Allow listening on both chains/atennas to get diversity.
* This might improve throughput slightly when far from the accesspoint,
apparently by allowing the firmware to listen on either of the two
antennas (if there are two, i.e. on 7260/7265/8260), whichever has
a better reception.
Taken-From: Linux iwlwifi
Tomohiro Kusumi [Wed, 12 Oct 2016 12:57:37 +0000 (21:57 +0900)]
sys/vfs/hammer: Use HAMMER_MAX_KEY for initial trunc_off values
ip->trunc_off variants are to be compared with node elm's base.key,
so it should use HAMMER_MAX_KEY on initialization instead of
0x7FFFFFFFFFFFFFFFLL.
Tomohiro Kusumi [Tue, 11 Oct 2016 15:05:19 +0000 (00:05 +0900)]
sys/vfs/hammer: Add HAMMER_MIN|MAX_ONDISK_LOCALIZATION
Use | for HAMMER_MAX_ONDISK_LOCALIZATION, but not +.
Also see
7e52af60.
Tomohiro Kusumi [Tue, 11 Oct 2016 14:40:45 +0000 (23:40 +0900)]
sys/vfs/hammer: Use MIN/MAX macros for B-Tree node's elm base
Leave obj_type with 0 instead of using HAMMER_OBJTYPE_UNKNOWN,
since it's not intended to be unknown, but literally intended
to be a value of 0.
François Tigeot [Tue, 11 Oct 2016 07:26:14 +0000 (09:26 +0200)]
drm/linux: Add sign_extend64()
François Tigeot [Tue, 11 Oct 2016 07:04:25 +0000 (09:04 +0200)]
drm/linux: Add kasprintf and kvasprintf()
François Tigeot [Tue, 11 Oct 2016 06:41:04 +0000 (08:41 +0200)]
drm/linux: Add kstrtouint()
Inspired from: FreeBSD
Matthew Dillon [Mon, 10 Oct 2016 17:22:12 +0000 (10:22 -0700)]
vkernel - Fix FP corruption in VMX/EPT mode
* Properly invalidating the EPT TLB cache when it potentially becomes
stale.
* When destroying a VMX context, make sure the EPT TLB cache has been
invalidated for that context as a safety.
* Fix a bug in vmx_vminit() where it was losing track of the currently
loaded VMX.
* Setup the VMX to clear the host CR0_TS, and the host makes sure the FP
state is correct prior to vmlaunch.
* Make sure the GUEST_CR0's CR0_TS reflects the vkernel says it should
reflect.
* The vkernel has a choice of asking the guest user context to #NM fault
on FP use or not. Both mechanics work, but its probably better for it
to ensure that the FP state is valid and tell the user context to not
fault. However, this commit makes sure that both mechanics work.
* Document why we shouldn't vmclear the old current context when loading
a different context.
* Cleanup some of the vkernel's pmap handling. This isn't perfect and
probably needs to be rewritten (we need a more powerful guest pmap
adjustment system call to properly synchronize changes). For now
we try to avoid races against detecting the (M)odified flag by clearing
the RW flag first.
Matthew Dillon [Sun, 9 Oct 2016 23:41:17 +0000 (16:41 -0700)]
vkernel - Add COW image capability
* Add a copy-on-write disk image feature. This allows a vkernel
to mount a disk image RO or R+W but does not try to write changes
back to the image file.
This allows multiple vkernel instances to use the same image
file.
* Note that when the vkernel operates on an image in this mode,
modifications will eat up system memory and swap, so the user
should be cognizant of the use-case. Still, the flexiblity of
being able to mount the image R+W should not be underestimated.
Imre Vadász [Sun, 9 Oct 2016 19:45:16 +0000 (21:45 +0200)]
iwm.4: Mention iwm7265Dfw firmware. Update DIAGNOSTICS section a bit.
Imre Vadász [Sat, 23 Jul 2016 13:54:50 +0000 (15:54 +0200)]
if_iwm - Add basic powermanagement support via ifconfig wlan0 powersave.
* The DEVICE_POWER_FLAGS_CAM_MSK flag was removed in the upstream iwlwifi
in Linux commit
ceef91c89480dd18bb3ac51e91280a233d0ca41f.
* Add sc_ps_disabled flag to struct iwm_softc, which corresponds to
mvm->ps_disabled in struct iwl_mvm in Linux iwlwifi.
* Adds a hw.iwm.power_scheme tunable which corresponds to the power_scheme
module parameter in Linux iwlwifi. Set this to 1 for completely
disabling power management, 2 (default) for balanced powermanagement,
and 3 for lowerpower mode (which does dtim period skipping).
* Imports the constants.h file from iwlwifi as if_iwm_constants.h.
* This doesn't allow changing the powermanagement setting while connected,
also one can only choose between enabled and disabled powersaving with
ifconfig (so switching between balanced and low-power mode requires
rebooting to change the tunable).
* After any changes to powermanagement (i.e. "ifconfig wlan0 powersave" to
enable powermanagement, or "ifconfig wlan0 -powersave" for disabling
powermanagement), one has to disconnect and reconnect to the accespoint
for the change to take effect.
Taken-From: Linux iwlwifi
Tomohiro Kusumi [Sun, 9 Oct 2016 09:22:18 +0000 (18:22 +0900)]
sys/dev/disk/dm: Make dmdebug() print __func__
and fix/remove dmdebug calls that could be cleaned up or simplified.
Tomohiro Kusumi [Sun, 9 Oct 2016 07:18:59 +0000 (16:18 +0900)]
sys/vfs/hammer: Mention ondisk fields are in little endian
Explicitly mention HAMMER1 uses little endian for ondisk structures,
but not host endian (which is little on currently supported arch).
This is easier because adding big endian support doesn't break
anything at least on DragonFly/x86_64.
This doesn't affect anything/anyone in reality because HAMMER1 on
DragonFly is the only HAMMER1 implementation (that I know of other
than the read only FUSE port).
Tomohiro Kusumi [Sat, 8 Oct 2016 07:22:34 +0000 (16:22 +0900)]
sbin/hammer: Remove getdir()
Using dirname(3) with a santiy check for trailing / on creating PFS
does the same as getdir(), so use a library rather than its own.
No one uses getdir() anymore, so it can be removed.
Tomohiro Kusumi [Sat, 8 Oct 2016 07:12:17 +0000 (16:12 +0900)]
sbin/hammer: Cleanup getpfs()/scanpfsid()
No functional difference.
Tomohiro Kusumi [Wed, 21 Sep 2016 22:04:18 +0000 (07:04 +0900)]
sbin/hammer: Partly revert
8b640662 in 2014
I guess it makes sense that
8b640662 silently ignores trailing / and
do destructive operations to a filesystem, since there isn't really
anything one can do by specifying a root inode (directory) of a PFS,
but it shouldn't silently modify (not ignore) the user input.
A command like pfs-status uses that input after calling getpfs(),
so the program should not silently modify the string and print it
as if that was the user input, because it's not. It should also avoid
modifying argv[i] even if it's technically possible (make it shorter).
It's also better to do it based on lstat(2) result than unconditionally
checking a trailing / based on assumption that a symlink that ends
with / is a directory (which is right if it actually exists).
The only case behaves differently from
8b640662 is when there is a
slave PFS that hasn't been synced with a master PFS yet after created.
In this case it fails to run commands using path arg with trailing /,
because lstat(2) mentioned above in scanpfsid() fails due to lack of
the root inode. This occurs by design of HAMMER, but not a tool issue.
Above won't be a problem either because the scenario mentioned in
8b640662 is mostly about auto completion by a shell, while in this
case the shell just recognizes it as a broken link. The tool doesn't
need to assist everything that users would potentially do based on
assumption that may be right or wrong.
Also see http://lists.dragonflybsd.org/pipermail/users/2015-April/228012.html
-----
Actually I even suspect reports by users and
8b640662 were wrong
in the first place, because one would still get the same PFS id from
both "symlink" and "symlink/" when "symlink" is a symlink to PFS.
"symlink" can extract the PFS id directly from "@@.." format, but
"symlink/" eventually results the same via hammer_pfs_autodetect().
In fact I could pfs-upgrade/pfs-downgrade by reverting whole
8b640662.
Having said that, it still needs to strip trailing / on pfs-destroy
otherwise it won't be able to remove a symlink to PFS after it gets
destroyed.
Tomohiro Kusumi [Sat, 8 Oct 2016 07:31:53 +0000 (16:31 +0900)]
sbin/hammer: Use dirname(3) instead of getdir() in scanpfsid()
scanpfsid() can use dirname(3) for fd instead of getdir().
It's actually better to use dirname(3) which strips trailing /
first and then return a pointer to directory path, instead of
getdir() which only strips trailing / if any, because that's
what scanpfsid() is trying to do with fd.
If path is
path/to/pfs -> @@...
scanpfsid() wants to open("path/to"), and if path is
path/to/pfs/
which is @@.. itself if the PFS is not destroyed,
scanpfsid() still wants to open("path/to").
The second one doesn't occur since
8b640662 changed getpfs()
to strip trailing / before calling scanpfsid().
However the next commit is going to partly revert
8b640662,
and this change makes the next 2 commits easier to implement,
while being logically separated topic.
Matthew Dillon [Sat, 8 Oct 2016 23:57:16 +0000 (16:57 -0700)]
kernel - Refactor VMX code
* Refactor the VMX code to use all three VMM states available to use
instead of two. The three states available are:
active and current (VMPTRLD)
active not current (replaced by some other context being VMPTRLD'd)
inactive not current (VMCLEAR)
In short, there is no need to VMCLEAR the current context when activating
another via VMPTRLD, doing so greatly reduces performance. VMCLEAR is
only really needed when a context is being destroyed or being moved to
another cpu.
* Also fixes a few bugs along the way.
* Live loop in vmx_vmrun() when necessary, otherwise we wind up with serious
problems synchronizing IPIs. The thread will still be subject to its
process priority.
Imre Vadász [Sat, 8 Oct 2016 21:56:42 +0000 (23:56 +0200)]
if_iwm - Add iwm_nic_unlock() calls missing from previous commit.
Imre Vadász [Sat, 8 Oct 2016 21:06:50 +0000 (23:06 +0200)]
if_iwm - iwm_{read,write}_prph() don't grab the nic lock in iwm themselves.
* Fix a couple of cases where the nic lock ended up not being grabbed
during an iwm_read_prph() or iwm_write_prph().
Sascha Wildner [Sat, 8 Oct 2016 06:06:42 +0000 (08:06 +0200)]
kernel/iwm: Fix building into the kernel without IWM_DEBUG.
Sascha Wildner [Sat, 8 Oct 2016 06:06:03 +0000 (08:06 +0200)]
kernel/iwm: Allow IWM_DEBUG to come in via the kernel config too.
Previously, it was only ever set for the module.