Tomohiro Kusumi [Mon, 6 Jul 2015 10:07:48 +0000 (19:07 +0900)]
sys/vfs/hammer: Cleanup HAMMER_ENCODE_XXX() macros
- Define HAMMER_ENCODE_XXX() macros using HAMMER_ENCODE()
which is a generic version of these.
Tomohiro Kusumi [Mon, 6 Jul 2015 09:57:29 +0000 (18:57 +0900)]
sys/vfs/hammer: Define and use HAMMER_ENCODE() on blockmap formatting
- Add HAMMER_ENCODE() which is a generic version of existing
macros HAMMER_ENCODE_{RAW_VOLUME|RAW_BUFFER|FREEMAP}().
- Use HAMMER_ENCODE() in format_blockmap() in userspace when newfs
initializes blockmap offsets. This is better in the sense that
the macro hides internal of hammer's storage address space format,
and gets rid of bitwise operations that make assumption on format.
Tomohiro Kusumi [Mon, 6 Jul 2015 09:15:15 +0000 (18:15 +0900)]
sbin/hammer: Remove panic()
- hammer userspace mostly uses err()/errx() and panic()
(which is only used once) can be replaced using these.
Sepherosa Ziehau [Sat, 4 Jul 2015 15:27:15 +0000 (23:27 +0800)]
powerd: Staticize global variables
Sepherosa Ziehau [Sat, 4 Jul 2015 15:20:52 +0000 (23:20 +0800)]
powerd: Merge TotalCpus and NCpus
Sepherosa Ziehau [Sat, 4 Jul 2015 15:13:34 +0000 (23:13 +0800)]
powerd.8: List only related kernel modules
Sepherosa Ziehau [Sat, 4 Jul 2015 14:41:43 +0000 (22:41 +0800)]
powerd: Remove unused global variable CpuCount[]
Sepherosa Ziehau [Thu, 2 Jul 2015 14:55:01 +0000 (22:55 +0800)]
x86_64: Add per-cpu mwait CX setting
Tomohiro Kusumi [Sat, 4 Jul 2015 14:16:46 +0000 (23:16 +0900)]
sbin/hammer: Remove test_volume()
- get_volume() and test_volume() are basically the same. The only
difference is get_volume() calls errx() when the volume is not
found while test_volume() returns NULL.
- The caller of test_volume() calls errx() when NULL is returned,
so test_volume() can be replaced using get_volume().
- Also note that if the volume instance is not found, there is
basically no way that the hammer userspace command can proceed
the rest of the command (similar to losing superblock in UFS).
John Marino [Sun, 5 Jul 2015 15:51:01 +0000 (17:51 +0200)]
gcc50 c++: Restore C99 after change to __LONG_LONG_SUPPORTED
The recent correction of __LONG_LONG_SUPPORTED had the undesirable
side of effect of breaking the base and dports GCC support for c++.
This patch restores it (comes from FreeBSD config), and will have to
be pushed upstream to GCC. All GCC without it (dports) now has
missing c++ functions (e.g. wcstoll, wcstoull)
Sascha Wildner [Sun, 5 Jul 2015 09:46:06 +0000 (11:46 +0200)]
<net/bpf.h>: Include <sys/param.h> for roundup2().
Reported-by: marino
Sascha Wildner [Sun, 5 Jul 2015 00:43:28 +0000 (02:43 +0200)]
kernel/uaudio: Fix weird indent.
Sascha Wildner [Sat, 4 Jul 2015 23:04:35 +0000 (01:04 +0200)]
<sys/socket.h>: Whitespace cleanup.
Sascha Wildner [Sat, 4 Jul 2015 23:00:50 +0000 (01:00 +0200)]
Remove duplicate includes in <iconv.h> and <sys/socket.h>.
François Tigeot [Sat, 4 Jul 2015 12:58:01 +0000 (14:58 +0200)]
drm/linux: Fix WARN_ONCE()
Sascha Wildner [Sat, 4 Jul 2015 11:27:26 +0000 (13:27 +0200)]
Use roundup2() from <sys/param.h> in a number of places.
John Marino [Sat, 4 Jul 2015 08:59:39 +0000 (10:59 +0200)]
Makefile.inc: Add kernel install complete message
All the targets that have an "starting" message also have a corresponding
"complete" message except the installkernel target. This is particularly
obvious when checking logs to see if kernel successfully installed, so
I'm adding that message here.
Sascha Wildner [Fri, 3 Jul 2015 23:30:00 +0000 (01:30 +0200)]
i386 removal, part 13/x: Some cleanup after pc32 platform removal.
Sascha Wildner [Fri, 3 Jul 2015 23:16:34 +0000 (01:16 +0200)]
i386 removal, part 12/x: Remove sys/platform/pc32.
Sascha Wildner [Fri, 3 Jul 2015 23:13:32 +0000 (01:13 +0200)]
i386 removal, part 11/x: Remove wrong machine/ setup in the boot Makefiles.
Several of the link targets were pointing to ${.CURDIR}/../../i386/...
which doesn't exist. This is from times when our platform-specific
headers were differently laid out and it was never noticed that after
some point, it was no longer needed.
In fact, sys/boot builds properly without all this, because:
* -m32 ensures that the i386 platform is targeted by the compiler
* Various (needed) __i386__ checks in sys/cpu/x86_64/include ensure that
even on x86_64, the basic types have the right sizes.
Confirmed by binary compare of /boot without and with the patch.
Sascha Wildner [Fri, 3 Jul 2015 23:01:40 +0000 (01:01 +0200)]
i386 removal, part 10/x: Fixes in sys/boot to compile without platform/pc32.
John Marino [Sat, 4 Jul 2015 07:29:50 +0000 (09:29 +0200)]
iconv sync 11/11: Use macro like FreeBSD (1 file) to eliminate diff
John Marino [Sat, 4 Jul 2015 07:15:01 +0000 (09:15 +0200)]
iconv sync 10/x: FreeBSD SVN 258316 + 283406 + 283418
SVN 258316: Bug fixes in iconv(3) UTF-7 support.
- Add ' to the list of directly encoded characters and * to the list of
optionally directly encoded characters as per RFC 2152.
- In _citrus_UTF7_mbtoutf16 on end of input when the next output character
has only been partially decoded, save a copy of the buffer of input
characters (not just its length). On the next call with more input
characters this buffer is reprocessed together with the new input to
form a fully decoded output character.
- At the end of a base64 encoded sequence fully discard '-' (BASE64_OUT)
by decrementing psenc->chlen and i. This is needed to make room in
psenc->ch (input buffer) in case the next input character starts a new
base64 encoded sequence. And also, if this is the end of input and no
output character can be returned, this brings the encoder in the initial
state as indicated by _citrus_UTF7_stdenc_get_state_desc_generic which
is used by the caller to distinguish between no output and partial
output.
- In _citrus_UTF7_mbrtowc_priv pass the s parameter (input pointer)
directly to _citrus_UTF7_mbtoutf16 instead of a copy (s0). This way s
is updated correctly in case of errors.
- In _citrus_UTF7_mbrtowc_priv when called with psenc->surrogate set
(previous call did not have enough input), retrieve the previously
decoded UTF-16 character fr
SVN 283406: Fix decoding of UTF-7 when a base64 encoded chunk appears at
the end of the input buffer.
_citrus_UTF7_mbtoutf16 stored the decoder state at the beginning so it
could restore this state on an incomplete character such that the next
call would restart the decoding. The problem was that "-" (end of base64
mode) at the end of a string was also treated as an incomplete character
but was also removed from the state buffer. So the initial state would be
restored (with base64 mode) and the next call would no longer see the "-"
so it continued in base64 mode.
This state saving/restoring isn't needed here. It's already handled
elsewhere (citrus_iconv_std.c:_citrus_iconv_std_iconv_convert) so just
remove it.
Also initialise *nresult.
SVN 283418: (continued)
When only 2 bytes can be read from a 4 byte UTF-16 character in a base64
encoded chunk of a UTF-7 string, treat that as an incomplete character and
return an error instead of a shift sequence and no error.
Also check that the low 2 bytes have a valid value.
John Marino [Sat, 4 Jul 2015 07:02:34 +0000 (09:02 +0200)]
iconv sync 9/x: FreeBSD SVN 260003
In libiconv i18n modules, surround the unused static
_citrus_xxx_pack_state() and _citrus_xxx_unpack_state() functions
with #if 0 blocks for now
John Marino [Fri, 3 Jul 2015 23:12:37 +0000 (01:12 +0200)]
iconv sync 8/x: FreeBSD SVN 281550
Remove the const qualifier from iconv(3) to comply with POSIX:
http://pubs.opengroup.org/onlinepubs/
9699919799/functions/iconv.html
Adjust all code that calls iconv
Note: The libcpp Makefile cflag was added because the host iconv.h
header was getting pulled in. This always should have been the
src version of iconv.h, but the error wasn't exposed until now.
John Marino [Fri, 3 Jul 2015 22:19:27 +0000 (00:19 +0200)]
iconv sync 7/x: FreeBSD SVN 281798
Fix improbable memory leak in _citrus_prop_read_str()
John Marino [Fri, 3 Jul 2015 22:14:00 +0000 (00:14 +0200)]
iconv sync 6/x: FreeBSD SVN 279404
Avoid lookup of CODESET aliases using unitialized path. We do not use
iconv.alias file so avoid using the vestiges of the code that do.
John Marino [Fri, 3 Jul 2015 22:06:32 +0000 (00:06 +0200)]
iconv sync 5/x: FreeBSD SVN 269529 + 269530 + 275805
- Add xrefs for other functions (in man page, then fix ordering)
- Fix incorrect type of "invalids" argument in __iconv() prototype
John Marino [Fri, 3 Jul 2015 22:00:45 +0000 (00:00 +0200)]
iconv sync 4/x: FreeBSD SVN 267438
iconv_open: initialize ci_ilseq_invalid field of _citrus_iconv_shared
struct after allocation with malloc
iconv_list: reduce a memory leak by copying strings only once
John Marino [Fri, 3 Jul 2015 21:54:27 +0000 (23:54 +0200)]
iconv sync 3/x: FreeBSD SVN 267437
Replace malloc+memset with calloc
John Marino [Fri, 3 Jul 2015 21:40:58 +0000 (23:40 +0200)]
iconv sync 2/x: FreeBSD SVN 263986
- In the libiconv module for ISO 2022, restore the original order of the
fields of a private struct such that variables of this type are
initialized correctly. This fixes conversion from ISO 2022
- In the libiconv module for EUC-TW, replace 2^cs with 1<<cs. This fixes
conversion from EUC-TW
- Synchronize iconv code from NetBSD. In most cases this only updates the
RCS id because the changes are already there or are NetBSD specific.
* citris_csmapper.c : Add a comment
* citrus_db_factory.c : Remove put16()
* citrus_iconv.c : Return EINVAL on error
* citrus_mapper.c : Return EINVAL on error
* citrus_stdenc.c : Return EINVAL on error
* citrus_memstream.c : Fix type of variable
* citrus_prop.h : Sync definition of _CITRUS_PROP_HINT_END
* citrus_mapper_std.c : Plug memory leak
Obtained from: NetBSD
John Marino [Fri, 3 Jul 2015 21:18:52 +0000 (23:18 +0200)]
iconv sync 1/x: FreeBSD SVN 262441
Consistently pass around context information using a simple pointer.
This fixes some dereferencing bugs in Chinese character set conversions.
Matthew Dillon [Sat, 4 Jul 2015 04:55:10 +0000 (21:55 -0700)]
tmpfs - Fix write-append/mmap-read race
* tmpfs_write() was not extending the target file atomically with the
write via a file node lock. Although the vnode is locked exclusively,
this only prevents read() and write() related races. lseek() and memory
mapped reads could still race.
* Adjust the NODE LOCK to cover the entire tmpfs_write() operation. This
fixes races against lseek() or stat() which would otherwise allow the
new post-append file size to be returned before the write completes.
-Matt
Reported-by: Ben Woolley <tautolog@gmail.com>
François Tigeot [Fri, 3 Jul 2015 20:21:00 +0000 (22:21 +0200)]
drm/i915: Sync with Linux 3.14 a tiny bit
The get_page() and set_pages_uc() Linux functions have been
implemented, we can use them.
Tomohiro Kusumi [Fri, 3 Jul 2015 13:51:23 +0000 (22:51 +0900)]
sbin/hammer: Use const char* for string literal args
Tomohiro Kusumi [Fri, 3 Jul 2015 13:50:45 +0000 (22:50 +0900)]
sys/vfs/hammer: Change "xxx-map" to "xxxmap"
- "xxxmap"s seem to be canonical terms rather than "xxx-map"s
where "xxx"s are name of zones.
- This commit only changes comments. Variable names and other
comments, etc already have "xxxmap".
Sascha Wildner [Fri, 3 Jul 2015 12:35:07 +0000 (14:35 +0200)]
<sys/lock.h>: Fix comment typo.
Submitted-by: Yellow Rabbit <yrabbit@sdf.lonestar.org>
Tomohiro Kusumi [Thu, 2 Jul 2015 14:41:06 +0000 (23:41 +0900)]
sys/vfs/hammer: Change "bigblock" to "big-block"
- There are(or were) several terms for 8MB chunk, for example
"big-block", "bigblock", "big block", "large-block", etc but
"big-block" seems to be the canonical term.
- Changes are mostly comments and some in printf and hammer(8).
Variable names (e.g. xxx_bigblock_xxx) remain unchanged.
- The official design document as well as much of the existing
code (excluding variable and macro names) use "big-block".
https://www.dragonflybsd.org/hammer/hammer.pdf
- Also see
e04ee2de and the previous commit.
Tomohiro Kusumi [Thu, 2 Jul 2015 14:08:07 +0000 (23:08 +0900)]
sys/vfs/hammer: Change "big block" to "big-block"
- This word refers to 8MB chunk in hammer's blockmap layers,
not literally "big" "block".
- Changes are mostly comments and some in printf and hammer(8).
- The official design document as well as much of the existing
code (excluding variable and macro names) use "big-block".
https://www.dragonflybsd.org/hammer/hammer.pdf
- Also see
e04ee2de.
Tomohiro Kusumi [Thu, 2 Jul 2015 13:50:12 +0000 (22:50 +0900)]
sys/vfs/hammer: Revert one line from ccdea23
- This needed to be kept with "btree" zone.
- No binary diff since these two macros are the same,
and also this code has been #if0'd.
Matthew Dillon [Wed, 1 Jul 2015 17:07:21 +0000 (10:07 -0700)]
kernel - Improve basic entropy collector
* Faster time-buffered period in first 120 seconds after boot to help seed
the generator. This helps when other sources such as RDRAND are not available.
* Use a systimer instead of tsleep to make the 'random' loop delay more random.
* Do a better job incorporating interrupt randomness with a combination of
counter and low TSC bits.
* Get rid of the double-time-buffering restrictions (one in the thread, one in
NANOUP_EVENT(). Also incorporate the low TSC bits in different bit positions
to try to avoid XOR deconstruction against tv_nsec.
Reported-by: tuxillo
Sascha Wildner [Tue, 30 Jun 2015 16:41:20 +0000 (18:41 +0200)]
kernel/acpi_thermal: Fix comment typo.
Sascha Wildner [Wed, 1 Jul 2015 09:35:08 +0000 (11:35 +0200)]
bsd-family-tree: Sync with FreeBSD (for DragonFly 4.2.0).
Matthew Dillon [Tue, 30 Jun 2015 21:47:46 +0000 (14:47 -0700)]
syscons - reenable cursor
* Cursor was accidently disabled in a prior commit,
reenable it.
Reported-by: swildner
Matthew Dillon [Tue, 30 Jun 2015 19:46:02 +0000 (12:46 -0700)]
drm - Used fixed start address instead of vm_map_hint() in i915_gem_mmap_ioctl()
* We tried to fix i915_gem_mmap_ioctl() returning a valid 'NULL' address for
the requested mapping by passing the vm_map_hint() address into it. This
worked, but for reasons unknown wound up destabilizing X11 itself.
* So, instead of using 0, and instead of using vm_map_hint(), the 'hint'
starting address is now fixed at PAGE_SIZE (0x0000000000001000). And this
seems to solve the destabilization.
Matthew Dillon [Tue, 30 Jun 2015 06:37:20 +0000 (23:37 -0700)]
hammer2 - xop stabilization - part 2/many
* The xop mechanics do a quorum calculation, and will also do quorum
calculations for the inode cluster (not yet in). This calculation is
primarily based on the bref.modify_tid.
Each backend XOP now needs its own modify_tid to properly distinguish
between and eventually serialize dependent backend ops.
* Implement discrete modify_tid (mtid) reservations in
hammer2_xop_alloc(). Remove the old modify_tid reservation code that only
incremented it on a per-flush basis.
* The flush code will acquire a unique modify_tid which is greater than all
prior modify_tids in stage 2 (after it clears PREFLUSH).
* The hammer2_xop_alloc() code is too messy returning the union structure,
return a void * instead so the caller can just assign it to the correct
sub-union structure.
Matthew Dillon [Tue, 30 Jun 2015 06:36:00 +0000 (23:36 -0700)]
kernel - Add atomic_fetchadd_64()
* Add an atomic_fetchadd_64() function which hammer2 needs. This might
not be possible on e.g. 32-bit architectures for porting purposes but
for now I don't want to put a lock around the hammer2 functions that
need it.
Sepherosa Ziehau [Mon, 29 Jun 2015 14:54:47 +0000 (22:54 +0800)]
powerd: Support Intel Performance and Energy Bias Hint
According to <<Intel Software Developer's Manual>>
"... guide the hardware heuristic of power management features to favor
increasing dynamic performance or conserve energy consumption."
hint 0 - Max performance.
hint 15 - Max energy saving.
Matthew Dillon [Tue, 30 Jun 2015 00:10:49 +0000 (17:10 -0700)]
hammer2 - xop stabilization - part 1/many
* Create XOP threads on-demand, they are needed sometimes even on PFSs
which don't normally need it.
* For spmp roots, count PFSTYPE_SUPROOT as a master to simplify the cluster
validation code.
* Allow hammer2_inode_get() to only repoint part of an inode's cluster.
* Stabilize the slave synchronizer (a little, requires more).
* Remove more dead code.
Bill Yuan [Sun, 28 Jun 2015 10:30:43 +0000 (18:30 +0800)]
ipfw3: join same filters with or operator
example:
ipfw3 add allow icmp to <ip addr 1> or <ip addr 2> ...
Sascha Wildner [Mon, 29 Jun 2015 08:05:38 +0000 (10:05 +0200)]
Convert acpi_timer to using ACPICA functions.
Bill Yuan [Sat, 27 Jun 2015 13:38:17 +0000 (21:38 +0800)]
ipfw3: filter from/to supports me/any/subnet
usage:
ipfw3 add allow udp to any
ipfw3 add allow icmp from me
ipfw3 add allow icmp from 192.168.1.0/24
Sepherosa Ziehau [Sun, 28 Jun 2015 13:40:58 +0000 (21:40 +0800)]
powerd: Save usched mask for later use, e.g. by perf-energy bias
Sepherosa Ziehau [Sun, 28 Jun 2015 12:59:38 +0000 (20:59 +0800)]
cpumask: Add XORMASK
Sepherosa Ziehau [Sun, 28 Jun 2015 12:06:40 +0000 (20:06 +0800)]
acpi/pstate: Allow users to set _PDL
Reported-by: swildner
Sepherosa Ziehau [Sun, 28 Jun 2015 11:51:19 +0000 (19:51 +0800)]
acpi/pstate: Rename sysctl node available_bin to avail
And hide sysctl node available from sysctl list.
Matthew Dillon [Sun, 28 Jun 2015 23:49:48 +0000 (16:49 -0700)]
hammer2 - Refactor frontend part 17
This ends the major refactoring. All major cluster ops have been converted
to XOPs. The new XOP APIs will be stabilized in subsequent commits, issues
will primarily be slave scan deadlocks.
* Refactor the fsync and slave scan code to use the XOP interface.
* Cleanup hammer2_cluster.c, removing numerous functions which are no
longer used.
Sascha Wildner [Sun, 28 Jun 2015 22:26:26 +0000 (00:26 +0200)]
Fix some typos in variable names.
Sascha Wildner [Sun, 28 Jun 2015 12:48:43 +0000 (14:48 +0200)]
kernel/acpica: Simplify machine dependent initialization a bit.
Sascha Wildner [Sun, 28 Jun 2015 09:52:54 +0000 (11:52 +0200)]
dloader: Bump the max. line length in includes (e.g. loader.conf) to 1024.
Some ACPICA debugging settings (such as a lists of layers and levels for
debug.acpi.layer and debug.acpi.level, see acpi(4)) can easily get longer
than 256.
Another likely candidate are hw.acpi.{install,remove}_interface.
Sascha Wildner [Sat, 27 Jun 2015 19:25:19 +0000 (21:25 +0200)]
acpiexec(8): Fix a crash with the '!!' command.
Taken from the ACPICA repository.
ACPICA commit log:
commit
3715a26caadc1847940e2759490410748a433b59
Author: Robert Moore <Robert.Moore@intel.com>
Date: Wed Jun 24 08:42:44 2015 -0700
Debugger: Fix problem with the !! command.
Prevent infinite loop caused by incorrectly entering the !!
command into the history buffer. ACPICA BZ 1171.
Matthew Dillon [Sat, 27 Jun 2015 17:34:48 +0000 (10:34 -0700)]
hammer2 - Refactor frontend part 16/many
* Replace cluster ops with more frontend/backend XOPs. Use XOPs to
resolve the root inode, handle the hidden directory, create the
hidden directory, and to run the unlinkq in the hidden directory.
* Cleanup additional cluster ops that can use ip->meta instead.
* Simplify hammer2_inode_unlock(), it is no longer cluster-centric so
do not pass-in a cluster to unlock and drop.
* hammer2 show - dump pfs fields for the super-root
* hammer2 set* directives, change the HAMMER2IOC_INODE_SET API slightly.
* newfs_hammer2 - name the super root directory for show output convenience.
Tomohiro Kusumi [Sat, 27 Jun 2015 16:03:29 +0000 (01:03 +0900)]
sbin/hammer: Add hammer volume-blkdevs command
- Many of the low-level hammer commands (e.g. show, blockmap,
checkmap, etc) require -f option with blkdevs format argument,
however hammer command does not provide a way to confirm volumes
in blkdevs format. One often has to manually type devices with
colons to generate blkdevs format via volume-list command.
- hammer volume-blkdevs prints a list of volumes in blkdevs format
which is just a different version of volume-list but useful.
- hammer volume-list and volume-blkdevs are basically the same.
The only difference is volume-list uses '\n' for separator and
volume-blkdevs uses ':' for separator.
# hammer volume-list /HAMMER
/dev/da1
/dev/da2
/dev/da3
# hammer volume-blkdevs /HAMMER
/dev/da1:/dev/da2:/dev/da3
#
Sascha Wildner [Sat, 27 Jun 2015 17:04:59 +0000 (19:04 +0200)]
rtld(1): Use standard boolean type from <stdbool.h>.
Sascha Wildner [Sat, 27 Jun 2015 16:24:58 +0000 (18:24 +0200)]
kernel: Use 'normal' types (i.e., uint8_t instead of __uint8_t).
Sascha Wildner [Sat, 27 Jun 2015 07:17:48 +0000 (09:17 +0200)]
kernel/acpica: Change button type to int.
Even though it can have only two values, it's not really a variable meant
to be either true or false.
Sepherosa Ziehau [Sat, 27 Jun 2015 14:59:32 +0000 (22:59 +0800)]
powerd: Move TurboOpt test to the beginning of the condition list
Sepherosa Ziehau [Sat, 27 Jun 2015 14:22:36 +0000 (22:22 +0800)]
acpi/pstate: Evaluate _OST after successful _PPC processing
Sepherosa Ziehau [Sat, 27 Jun 2015 13:46:26 +0000 (21:46 +0800)]
acpi/pstate: Support _PDL, which limits number of usable P-States
Reported-by: swildner
Sepherosa Ziehau [Sat, 27 Jun 2015 13:10:47 +0000 (21:10 +0800)]
acpi/pstate: Fix _PPC support
Reported-by: swildner
Sepherosa Ziehau [Sat, 27 Jun 2015 08:39:07 +0000 (16:39 +0800)]
powerd: Make sure that all CPUs are found before we move on
Sepherosa Ziehau [Sat, 27 Jun 2015 08:16:41 +0000 (16:16 +0800)]
powerd: Utilize new acpi/pstate sysctl node for available frequencies
While I'm here, clean up CPU power domain frequency change.
Sepherosa Ziehau [Sat, 27 Jun 2015 04:13:20 +0000 (12:13 +0800)]
acpi/pstate: Add binary form for available CPU power domain frequencies
Sepherosa Ziehau [Fri, 26 Jun 2015 14:31:52 +0000 (22:31 +0800)]
powerd: Fix usched cpumask setting
CPU power domain does not necessarily contain CPUs, whose ids are
contiguous.
Sepherosa Ziehau [Fri, 26 Jun 2015 12:49:33 +0000 (20:49 +0800)]
powerd: Use linked list for CPU power domain
On some multiple CPU packages systems, CPU power domains' id could
be pretty sparse.
Sepherosa Ziehau [Thu, 25 Jun 2015 14:38:24 +0000 (22:38 +0800)]
powerd: Avoid magic number and nuke set-only CpuToDom
Matthew Dillon [Sat, 27 Jun 2015 02:23:45 +0000 (19:23 -0700)]
hammer2 - Refactor frontend part 15/many
* Cleanup pass, remove dead cluster API functions and other dead code.
* Cluster not completely gone yet, the synchronization code, a few ioctls,
and portions of vfsops are still using it.
Matthew Dillon [Sat, 27 Jun 2015 02:12:26 +0000 (19:12 -0700)]
hammer2 - Refactor frontend part 14/many
* Implement nlink, nremove, nrmdir, and nrename.
The hardlink handling in nlink and nremove were the most difficult
here, but the resulting design is actually cleaner than the complex
do-everything routines I had before.
* Remove the old hardlink handling clutter in the inode and cluster code.
Matthew Dillon [Fri, 26 Jun 2015 05:53:23 +0000 (22:53 -0700)]
hammer2 - Refactor frontend part 13/many
* Fix improper unlock in xop_scanlhc.
* Move the flush sequencing to the XOP backend.
* Make a kmalloc'd copy of the namecache name for backends still
in-progress when the frontend returns (we will need to do something
similar for bio data too).
* Do not pass the nch to hammer2_unlink_file(), pass more portable
integer state instead.
Matthew Dillon [Fri, 26 Jun 2015 21:05:40 +0000 (14:05 -0700)]
kernel - Fix syscons deadlock during panic
* The system console can deadlock during a panic if a cpu is stopped
while holding syscons_mtx.
* The new asynchronous update thread makes this problem more likely to occur.
* Fix the problem by recoding syscons_lock() to detect the panic or shutdown
condition and loop for up to 0.5 seconds trying to get the syscons_mtx.
If it is unable to acquire it, it reinitializes the mutex.
* We still have issues with VT switching away from X.
Sascha Wildner [Fri, 26 Jun 2015 16:09:30 +0000 (18:09 +0200)]
dump(8): Remove some old code.
Tomohiro Kusumi [Thu, 25 Jun 2015 20:21:46 +0000 (05:21 +0900)]
sys/vfs/hammer: Fix comments on zones
- zone-3 is also converted to zone-2 just like zone >= 8.
- "meta" is the actual name for zone-9 although meta means
the same as record.
Tomohiro Kusumi [Thu, 25 Jun 2015 19:06:04 +0000 (04:06 +0900)]
sys/vfs/hammer: Define HAMMER_ZONE2_MAPPED_INDEX for zone# assertions
- Define HAMMER_ZONE2_MAPPED_INDEX as zone-8(btree zone) index and
use this (instead of zone-8 index) to test if the given zone# is
is valid zone#. This makes better sense as "btree" zone itself
doesn't have much meaning with regards to checking zone# range
(e.g. meta zone or large/small data zones could have been defined
as zone-8 when hammer was designed and it would still work with
the non-btree-zone-8).
- This is similar to the way HAMMER_RECTYPE_ENTRY_START is defined
and used to lookup range of records.
Tomohiro Kusumi [Thu, 25 Jun 2015 18:45:47 +0000 (03:45 +0900)]
sbin/newfs_hammer: Cleanup zone index
- Use HAMMER_ZONE_BTREE_INDEX instead of 8.
Matthew Dillon [Fri, 26 Jun 2015 02:43:31 +0000 (19:43 -0700)]
hammer2 - Refactor frontend part 12/many
* Refactor the strategy_write code to use the XOP interface.
The new strategy_write uses the standard XOP interface to distribute
BIOs to XOP threads. As with the read code, this also tends to
distribute the compression across available cpus.
* Get rid of the dedicated BIO threading code (wthread* in hammer2_pfs).
Matthew Dillon [Thu, 25 Jun 2015 22:06:15 +0000 (15:06 -0700)]
hammer2 - Refactor frontend part 11/many
* Add XOPs for nresolve and most inode creation. The inode creation support
backs all inode-creation XOPs.
* Remove the on-stack hammer2_trans_t structure and remove the passing
of &trans to all modifying operations.
This methodology added significant list-management overhead and isn't
a good fit with the new frontend/backend mechanics. The removal also
simplifies the API for many support functions.
Replace with a temporary version that embeds a single, simplified
hammer2_trans_t in each hammer2_pfs_t. Ultimately we will need
to fan-out to per-node (per hammer2_mount_t) transaction control
in a way that doesn't stall-out the frontend.
François Tigeot [Thu, 25 Jun 2015 09:52:40 +0000 (11:52 +0200)]
rc.subr: Add common functions for run_rc_command
They are required by some ports rc scripts, the bind and haproxy
ports being the most obvious users.
Obtained-from: FreeBSD
Matthew Dillon [Wed, 24 Jun 2015 06:28:15 +0000 (23:28 -0700)]
hammer2 - Refactor frontend part 10/many
* New file hammer2_xops.c will contain the backend XOPs. In this
commit, just hammer2_xop_readdir() so far.
Matthew Dillon [Wed, 24 Jun 2015 06:01:54 +0000 (23:01 -0700)]
hammer2 - Refactor frontend part 9/many
* Create initial frontend/backend XOP infrastructure.
frontend:
hammer2_xop_alloc()
hammer2_xop_start()
... hammer2_xop_collect() loop ...
hammer2_xop_retire(xop, HAMMER2_XOPMASK_VOP)
backend:
(backend is called with the shared xop structure in separate
backend threads for each node belonging to the cluster appropriate
for the operation).
... issue chain calls as needed ...
... hammer2_xop_feed() ... (feed chains back to frontend)
hammer2_xop_feed(NULL) (feed NULL chain)
hammer2_xop_retire(xop, 1U << clindex)
The XOP contains a FIFO, allowing the backend to pipeline results when
appropriate (e.g. readdir). If a sequence of results are expected, the
backend should finish with a NULL chain. If not, the backend can just feed
back whatever is expected. Often this will just be the chain representing
the inode.
The frontend calls hammer2_xop_collect() to collect results from all the
backend nodes. The collect function handles quorum validation and
consolidates the results from a sufficient number of cluster nodes into
a single result for the frontend.
* The frontend can disconnect from the operation at any time in order to
be able to return a result, even if backend elements are still running.
This typically occurs when a sufficient number of nodes in the cluster
have responded to validate the quorum.
This also allows backend nodes to stall indefinitely without stalling the
frontend.
* Because frontend concurrency is lost due to the bulk of the work being
done by the backend, the hammer2 mount code will allocate ~16 or so
work threads per node to distribute potentially many frontend operations.
* Most frontend operations use existing cache layers to retains frontend
concurrency. Specifically, inode meta-data access and modifications,
and logical buffer cache operations (when cached), and cached vnodes
via the namecache. If the cache is not available, operations will
wind up using the VOP/XOP infrastructure, including buffer-cache
strategy routines (in an upcoming commit).
* Implement readdir() using the new infrastructure as an initial test.
* Fix an ip->meta synchronization bug related to hardlinks that was
introduced by the ip->meta local copy work.
Matthew Dillon [Wed, 24 Jun 2015 05:59:15 +0000 (22:59 -0700)]
kernel - Add zero-on-instantiate objcache ctor
* When creating kmalloc-backed pools, M_ZERO cannot be passed to
objcache_get() because the underlying kmalloc only occurs when
the objcache is not recycling a cached entry.
* Add a feature to the objcache whereby the allocation from backing
store is zero'd. The reuse case will not be zerod, so users of
this type of objcache must properly reset/cleanup fields before
disposing of the object.
* Used by HAMMER2.
Sepherosa Ziehau [Tue, 23 Jun 2015 15:32:52 +0000 (23:32 +0800)]
usched: Fix kern.usched_global_cpumask sysctl type.
Sepherosa Ziehau [Tue, 23 Jun 2015 14:56:15 +0000 (22:56 +0800)]
powerd: Make sure that cpu0 is in the usched global cpumask.
Sepherosa Ziehau [Tue, 23 Jun 2015 14:48:10 +0000 (22:48 +0800)]
powerd: Fix # of supported cpus
Sascha Wildner [Tue, 23 Jun 2015 21:24:02 +0000 (23:24 +0200)]
Sync zoneinfo database with tzdata2015e from ftp://ftp.iana.org/tz/releases
* africa: Morocco suspends DST from 2015-06-14 03:00 through
2015-07-19 02:00, not 06-13 and 07-18 as we had guessed.
(Thanks to Milamber.)
* northamerica: Assume Cayman Islands will observe DST starting next year,
using US rules. Although it isn't guaranteed, it is the most likely.
* southamerica: Typo fix.
Sascha Wildner [Tue, 23 Jun 2015 20:41:44 +0000 (22:41 +0200)]
i386 removal, part 9/x: Remove the 32 bit vkernel platform code.
Sascha Wildner [Mon, 22 Jun 2015 18:13:19 +0000 (20:13 +0200)]
i386 removal, part 8/x: Remove i386 DragonFly 1.2 compat kernel code.
Reclaim all the COMPAT_DF12 type syscalls since they never existed on
x86_64.
Sascha Wildner [Mon, 22 Jun 2015 17:59:42 +0000 (19:59 +0200)]
i386 removal, part 7/x: Remove the old i386 DragonFly 1.2 compat libraries.
Sascha Wildner [Mon, 22 Jun 2015 03:42:37 +0000 (05:42 +0200)]
i386 removal, part 6/x: Remove i386 parts of the ISA and PCI buses.
Sascha Wildner [Mon, 22 Jun 2015 03:41:49 +0000 (05:41 +0200)]
kernel: Include generic headers which will take care of platforms.
Sascha Wildner [Mon, 22 Jun 2015 03:01:38 +0000 (05:01 +0200)]
i386 removal, part 5/x: Remove i386 parts of the HighPoint RAID drivers.