Bring in a number of mps(4) driver fixes from LSI:
1. Fixed timeout specification for the msleep in mps_wait_command().
Added 30 second timeout for mps_wait_command() calls in mps_user.c.
2. Make sure we call mps_detach_user() from the kldunload path.
3. Raid Hotplug behavior change.
The driver now removes a volume when it goes to a failed state,
so we also need to add volume back to the OS when it goes to
opitimal/degraded/online from failed/missing.
Handle raid volume add and remove from the IR_Volume event.
4. Added some more debugging information.
5. Replace xpt_async(AC_LOST_DEVICE, path, NULL) with
mpssas_rescan_target().
This is to work around a panic in CAM that shows up when adding a
drive with a rescan and removing another device from the driver thread
with an AC_LOST_DEVICE async notification.
This problem was encountered in testing with the LSI sas2ircu utility,
which was used to create a RAID volume from physical disks. The driver
has to create the RAID volume target and remove the physical disk
targets, and triggered a panic in the process.
The CAM issue needs to be fully diagnosed and fixed, but this works
around the issue for now.
6. Fix some memory initialization issues in mps_free_command().
7. Resolve the "devq freeze forever" issue. This was caused by the
internal read capacity command issued in the non-head version of the
driver. When the command completed with an error, the driver wasn't
unfreezing thd device queue.
The version in head uses the CAM infrastructure for getting the read
capacity information, and therefore doesn't have the same issue.
8. Bump the version to 13.00.00.00-fbsd. (this is very close to LSI's
internal stable driver 13.00.00.00)
Submitted by: Kashyap Desai <Kashyap.Desai@lsi.com>
MFC after: 3 days
Luigi Rizzo [Wed, 8 Feb 2012 11:43:29 +0000 (11:43 +0000)]
- change the buffer size from a constant to a
TUNABLE variable (hw.netmap.buf_size) so we can experiment
with values different from 2048 which may give better cache performance.
- rearrange the memory allocation code so it will be easier
to replace it with a different implementation. The current code
relies on a single large contiguous chunk of memory obtained through
contigmalloc.
The new implementation (not committed yet) uses multiple
smaller chunks which are easier to fit in a fragmented address
space.
Brooks Davis [Wed, 8 Feb 2012 11:30:13 +0000 (11:30 +0000)]
eui64_aton and eui64_ntoa are actually the equivalent of ether_aton_r and
ether_nota_r and do not use static variables so remove the note copied
from ethers.3 saying they do.
Doug Barton [Wed, 8 Feb 2012 08:52:40 +0000 (08:52 +0000)]
As it stands right now, the default devfs rulesets are only loaded as a
side effect of something else using them. If they haven't been loaded
already but you want to use them, say for configuring a jail, you're out
of luck.
So add a knob to always load the default rulesets. While I'm here document
the other devfs_ knobs in rc.conf.5.
Jung-uk Kim [Tue, 7 Feb 2012 20:54:44 +0000 (20:54 +0000)]
- Give all clocks and timers on acpi0 the equal probing order.
- Increase probing order for ECDT table to match HID-based probing.
- Decrease probing order for HPET table to match HID-based probing.
- Decrease probing order for CPUs and system resources.
- Fix ACPI_DEV_BASE_ORDER to reflect the reality.
Kirk McKusick [Tue, 7 Feb 2012 20:43:28 +0000 (20:43 +0000)]
In the original days of BSD, a sync was issued on every filesystem
every 30 seconds. This spike in I/O caused the system to pause every
30 seconds which was quite annoying. So, the way that sync worked
was changed so that when a vnode was first dirtied, it was put on
a 30-second cleaning queue (see the syncer_workitem_pending queues
in kern/vfs_subr.c). If the file has not been written or deleted
after 30 seconds, the syncer pushes it out. As the syncer runs once
per second, dirty files are trickled out slowly over the 30-second
period instead of all at once by a call to sync(2).
The one drawback to this is that it does not cover the filesystem
metadata. To handle the metadata, vfs_allocate_syncvnode() is called
to create a "filesystem syncer vnode" at mount time which cycles
around the cleaning queue being sync'ed every 30 seconds. In the
original design, the only things it would sync for UFS were the
filesystem metadata: inode blocks, cylinder group bitmaps, and the
superblock (e.g., by VOP_FSYNC'ing devvp, the device vnode from
which the filesystem is mounted).
Somewhere in its path to integration with FreeBSD the flushing of
the filesystem syncer vnode got changed to sync every vnode associated
with the filesystem. The result of this change is to return to the
old filesystem-wide flush every 30-seconds behavior and makes the
whole 30-second delay per vnode useless.
This change goes back to the originally intended trickle out sync
behavior. Key to ensuring that all the intended semantics are
preserved (e.g., that all inode updates get flushed within a bounded
period of time) is that all inode modifications get pushed to their
corresponding inode blocks so that the metadata flush by the
filesystem syncer vnode gets them to the disk in a timely way.
Thanks to Konstantin Belousov (kib@) for doing the audit and commit
-r231122 which ensures that all of these updates are being made.
Marius Strobl [Tue, 7 Feb 2012 20:24:52 +0000 (20:24 +0000)]
Call bge_add_sysctls() early and especially before bge_can_use_msi() so
r230337 actually has a chance of working and doesn't always unconditionally
disable the use of MSIs.
Rick Macklem [Tue, 7 Feb 2012 16:32:43 +0000 (16:32 +0000)]
r228827 fixed a problem where copying of NFSv4 open credentials into
a credential structure would corrupt it. This happened when the
p argument was != NULL. However, I now realize that the copying of
open credentials should only happen for p == NULL, since that indicates
that it is a read-ahead or write-behind. This patch fixes this.
After this commit, r228827 could be reverted, but I think the code is
clearer and safer with the patch, so I am going to leave it in.
Without this patch, it was possible that a NFSv4 VOP_SETATTR() could have
changed the credentials of the caller. This would have happened if
the process doing the VOP_SETATTR() did not have the file open, but
some other process running as a different uid had the file open for writing
at the same time.
Sprinkle missed calls to asynchronous UFS_UPDATE() in attempt to
guarantee that all UFS inode metadata changes results in the dirtiness
of the inodeblock. Due to missed inodeblock updates, syncer was
required to fsync each mount point' vnode to guarantee periodic
metadata flush.
Don Lewis [Mon, 6 Feb 2012 21:50:11 +0000 (21:50 +0000)]
Improve sparse file handling when printing the block list for an inode by
not bailing out early when a hole is encountered in the direct block list.
Print NULL block pointers in the direct block list. Simplify the
code that prints the fragment count.
Dimitry Andric [Mon, 6 Feb 2012 21:35:11 +0000 (21:35 +0000)]
In usr.bin/rpcgen/rpc_main.c, use execvp(3) instead of execv(3), so
rpcgen will search the current PATH for the preprocessor. This makes it
possible to run a preprocessor built during the cross-tools stage of
buildworld.
Adrian Chadd [Mon, 6 Feb 2012 20:23:21 +0000 (20:23 +0000)]
Contribute some example code which demonstrates how to initialise the
radar parameters for the AR5416 and later NICs.
These parameters have been tested on the following NICs:
* AR5416
* AR9160
* AR9220
* AR9280
And yes, these will return radar pulse parameters and (for AR9160 and later)
radar FFT information as PHY errors.
This is again not enough to do radar detection, it's just here to faciliate
development and validation of radar detection algorithms.
The (pulse, not FFT) decoding code for AR5212, AR5416 and later NICs exist
in the HAL.
This code is disabled for now as generating radar PHY errors can quickly
cause issues in busy environment.s Some further debugging of the RX path
is needed.
Finally, these parameters are likely not useful for the AR5212 era NICs.
The madwifi-dfs branch should have suitable example parameters for the
11a era NICs.
Ed Schouten [Mon, 6 Feb 2012 18:52:40 +0000 (18:52 +0000)]
Add a `fix' for another whitespace bug.
If the sentence starts with a multiple of eight spaces, the sentence
should in almost all practical cases have started with tabs instead.
Replace these spaces by tabs.
Guy Helmer [Mon, 6 Feb 2012 14:11:24 +0000 (14:11 +0000)]
Using the O_CLOEXEC flag on open(2) caused the pidfile lock to be lost
when the child process execs daemon's target program thanks to flock(2)
semantics. So, we apparently have to leak the open pidfile's file
descriptor to keep the lock for the pidfile(3) functions to work properly.
Dimitry Andric [Mon, 6 Feb 2012 12:03:21 +0000 (12:03 +0000)]
Let rpcgen(1) support an environment variable RPCGEN_CPP to find the C
preprocessor to run. Previously, it always ran /usr/bin/cpp, unless you
used the -Y option, and even then you could not set the basename. It
also attempted to run /usr/ccs/lib/cpp for SVR4 compatibility, but this
is obsolete, and has been removed.
Note that setting RPCGEN_CPP to a command with arguments is supported,
though the command line parsing is simplistic. However, setting it to
e.g. "gcc46 -E" or "clang -E" will lead to problems, because both gcc
and clang in -E mode will consider files with unknown extensions (such
as .x) as object files, and attempt to link them.
This could be worked around by also adding "-x c", but it is much safer
to set RPCGEN_CPP to e.g. "cpp46" or "clang-cpp" instead.
JNEWBLK dependency may legitimately appear on the buf dependency
list. If softdep_sync_buf() discovers such dependency, it should do
nothing, which is safe as it is only waiting on the parent buffer to
be written, so it can be removed.
Current implementations of sync(2) and syncer vnode fsync() VOP uses
mnt_noasync counter to temporary remove MNTK_ASYNC mount option, which
is needed to guarantee a synchronous completion of the initiated i/o
before syscall or VOP return. Global removal of MNTK_ASYNC option is
harmful because not only i/o started from corresponding thread becomes
synchronous, but all i/o is synchronous on the filesystem which is
initiated during sync(2) or syncer activity.
Instead of removing MNTK_ASYNC from mnt_kern_flag, provide a local
thread flag to disable async i/o for current thread only. Use the
opportunity to move DOINGASYNC() macro into sys/vnode.h and
consistently use it through places which tested for MNTK_ASYNC.
Some testing demonstrated 60-70% improvements in run time for the
metadata-intensive operations on async-mounted UFS volumes, but still
with great deviation due to other reasons.
These tools are mainly useful for people that want to manipulate llvm
bitcode (.bc) and llvm assembly language (.ll) files, or want to tinker
with llvm and clang themselves.
Andreas Tobler [Sun, 5 Feb 2012 20:04:43 +0000 (20:04 +0000)]
Replace the assembler macro WEAK_ALIAS with a new macro WEAK_REFERENCE which
has the same API as __weak_reference(). Give 'x' in SYS.h a more meaningful
name.
Gleb Smirnoff [Sun, 5 Feb 2012 16:53:02 +0000 (16:53 +0000)]
Add new socket options: TCP_KEEPINIT, TCP_KEEPIDLE, TCP_KEEPINTVL and
TCP_KEEPCNT, that allow to control initial timeout, idle time, idle
re-send interval and idle send count on a per-socket basis.
Mikolaj Golub [Sun, 5 Feb 2012 15:23:32 +0000 (15:23 +0000)]
If a local write request is from the synchronization thread, when it
is synchronizing data that is out of date on the local component, we
should not send G_GATE_CMD_DONE acknowledge to the kernel.
This fixes the issue, observed in async mode, when on synchronization
from the remote component the worker terminated with "G_GATE_CMD_DONE
failed" error.
Mikolaj Golub [Sun, 5 Feb 2012 15:21:08 +0000 (15:21 +0000)]
Fix the regression introduced in r226859: if the local component is
out of date BIO_READ requests got lost instead of being sent to the
remote component.
Add support for bus_adjust_resource() on all PowerPC/AIM PCI bridges. With
this change, NEW_PCIB appears to work without incident at least on a G5
iMac. More testing will be required before it is turned on in GENERIC.
Jilles Tjoelker [Sat, 4 Feb 2012 23:12:14 +0000 (23:12 +0000)]
sh: Use vfork in a few common cases.
This uses vfork() for simple commands and command substitutions containing a
single simple command, invoking an external program under certain conditions
(no redirections or variable assignments, non-interactive shell, no job
control). These restrictions limit the amount of code executed in a vforked
child.
There is a large speedup (for example 35%) in microbenchmarks. The
difference in buildkernel is smaller (for example 0.5%) but still
statistically significant. See
http://lists.freebsd.org/pipermail/freebsd-hackers/2012-January/037581.html
for some numbers.
The use of vfork() can be disabled by setting a variable named
SH_DISABLE_VFORK.
After multiple requests for back out, after multiple promisses from adrian@
to fix it (without fixing it), after making one of my servers unbootable,
after now making also my laptop unbootable and after running out of patiance
back out r229800 until better solution is found.
Unify OF PCI infrastructure, including changing from parsing the device
tree based on heuristics to parsing it based on the spec. This should also
lay the foundation for NEW_PCIB on PowerPC.
Ryan Stone [Sat, 4 Feb 2012 16:49:29 +0000 (16:49 +0000)]
Whenever a new kernel thread is spawned, explicitly clear any CPU affinity
set on the new thread. This prevents the thread from inadvertently
inheriting affinity from a random sibling.