Add __cacheline_aligned and __read_mostly annotations.

These annotations help to mitigate false sharing on multiprocessor
systems.

Variables annotated with __cacheline_aligned are placed into the
.data.cacheline_aligned section in the kernel. Each item in this
section is aligned on a cachline boundary - this avoids false
sharing. Highly contended global locks are a good candidate for
__cacheline_aligned annotation.

Variables annotated with __read_mostly are packed together tightly
into a .data.read_mostly section in the kernel. The idea here is that
we can pack infrequently modified data items into a cacheline and
avoid having to purge the cache, which would happen if read mostly
data and write mostly data shared a cachline. Initialisation variables
are a prime candiate for __read_mostly annotations.

(mjf)

After a recent WARNS= bump on ld.elf_so it fails to build on hppa,
complaining that we're "discarding qualifier from pointer target
type".

Use const char * instead of caddr_t to avoid the warning and get
distribution building on hppa again.

(mjf)

u_intNN_t -> uintNN_t

(mjf)

Add an exmaple of unmounting all file systems apart from NFS and MFS
e.g,

umount -a -t nonfs,mfs

This makes it clear that the "no prefix" only needs to be on the first
file-system argument to -t. The mount(8) man page has a clear example
of this but for some reason umount(8) was missing one.

(mjf)

Add PA-RISC 2.0 PIM support.

Fix the HPMC exception handler so that, if we're running on a PA-RISC
2.0 machine, we use the 64-bit PIM data structures.

There was also a bug in the HPMC exception handler that stopped output
being written to the console after we'd taken the exception. We need
to make a PDC_IO pdc call to reset I/O.

(mjf)

Add a license with my name on it. Suggested by skrll@

(mjf)

Get audio capture fully working with harmony(4) and fix PR/35239.

skrll@ reports that with this patch he can record and playback audio.
Suprisingly the bug was actually with playback and not with capture.
When not capturing or playing audio we write to or read from "empty
buffers", the problem was that playback and capture were using the same
buffer and the playback code managed to pick up a bit of data that was
written from the capture code.

(mjf)

Continue my crusade - queueing -> queuing

(mjf)

Queueing -> Queuing

(mjf)

Typo, "Queueing" -> "Queuing"

(mjf)

Typo and grammar fixes.

- s/though/through/ appropriately.
- autoconf(4) doesn't exist, autoconf(9) is the correct reference.

(mjf)

Regen.

(mjf)

Add Initio INIC-1622 SATA Controller.

(mjf)

Doh, I somehow didn't commit these two changes initially.

Set pwr_sw_control by hand instead of calling the wrapper function
because the wrapper may invoke the cold_hook func, which is undesirable
at attach time. Remove the check for cold != 0 in pwr_sw_ctrl(), we
know when it is called (only when the machine is warm).

(mjf)

Merge our power code and power(4) from OpenBSD. I removed a lot of global
state and structured it to look more like normal device drivers. power0
attaches at mainbus.

We retain sysctl compatibility.

Thanks to Nick Hudson for reviewing and testing.

(mjf)

Document the Processor-Dependent Code firmware driver. From OpenBSD.

(mjf)

- Remove needless assignments to 'nca' seeing as it gets zero'd out before
  being used again.
- Update a comment to explain that we attach all CPUs and memory
  controllers in one step - we don't attach the main CPU first and then
  additional CPUs.

(mjf)

Make sure the pipline has been completely flushed after executing an rfi
instruction. Without this I was being hit with an "Illegal instruction"
trap.

(mjf)

mscp(4) was being a bad monkey! It's not OK to call config_found() from
interrupt context, as my vax kernel with DIAGNOSTIC told me. Instead, we
defer the work and get it to run in thread context via a workqueue(9).

Thanks to matt@ for the review.

(mjf)

It seems that "fattr.h" moved and what I should now include is
"mountprog.h" to get a_gid() and a_uid().

(mjf)

Sync with HEAD.

(mjf)

- Fix a CVS merge breakage
- Catch up with some driver changes

(mjf)

Fix printf format for suseconds_t

(mjf)

Catch up with shlib major bump for all arm platforms.

(mjf)

Sync with HEAD.

(mjf)

Sync with HEAD.

(mjf)

- Separate the COMPAT_13 and COMPAT_16 code into their own files and
  compile them only when the relevant config options are set.

- Provide a sendsig_sigcontext() implementation using the existing vax
  signal code; it already does the correct thing.

(mjf)

Add missing dollar sign ($) for $NetBSD$ tag.

(mjf)

The .if conditional crosses a line boundary, so add a backslash.

This fixes the kernel build for hppa.

(mjf)

Back out my previous change. PMAP_NC is blatantly used, which means I
just broke the build.

(mjf)

Delete PMAP_NC because it's never used. To mark a vmpage as uncacheable
we use pmap_pvh_attrs(TLB_UNCACHEABLE) instead.

(mjf)

Bah! Fix my previous commit that broke the build. I should have converted
the cast I just removed to uint32_t.

(mjf)

Don't cast 'pve->pv_va' to int, seeing as its type is vaddr_t (uint32_t).

(mjf)

Fix a typo I saw when turning on TRAPDEBUG. struct pmap doesn't have
a pmap_space member, it should be pm_space.

(mjf)

Fix debug code that had bitrotted. This allows the maple(4) driver to
be compiled with MAPLE_DEBUG defined.

o line 798: variable buf shadows previous local
o line 1425: event_thread is no longer a struct proc but now is
    a lwp_t. Use the correct p_pid member.

(mjf)

Sync with HEAD.

(mjf)

Sync with HEAD.

(mjf)

Sync with HEAD.

(mjf)

Sync with HEAD.

(mjf)

Sync with HEAD.

(mjf)

Almost fix capturing audio with harmony(4).

There is a still a bug whereby a chunk of previously recorded data may
appear in the recorded audio file. This goes part way to fixing PR/35239

(mjf)

Fix pseye_start_transfer() to return the return value of kthread_create()
instead of always returning 0.

(mjf)

If it is not impossible to internalize the proplib dictionary from
devsfd.conf then set errno to EINVAL so that the user has some idea what
went wrong.

(mjf)

DCTL_NEW_MOUNT events are also placed onto a separate "mount event" list.
They are removed from this list when a DCTL_UNMOUNT event for the
corresponding mount gets sent to dctl(4).

This allows devfsd(8) to always know about devfs mounts on a system,
even if one was mounted whilst devfsd was not running.

(mjf)

Add -s option which allows devfsd to run in the foreground and not wait
for new events. When there are no more events to be handled, it exits.

(mjf)

struct componentname has no cn_lwp member anymore. Also update the data
types for the other struct members so they actually reflect sys/sys/namei.h

(mjf)

dctl:
- Make dctl see aware of device driver names being added and removed so
  that it can signal devfsd(8), which in turn will request device node
  creation of deletion from a particular devfs mount.

- Each device that registers the need for a device node also passes on
  some info about what sort of device they are. We now package this up
  in the dctl messaging framework to let devfsd(8) know.

- Use a dev_t for the cookie for each device that dctl(4) knows about. This
  was changed from a device_t because some devices (zero(4)) don't have
  corresponding device_t's.

- dctl mount functions have grown a visibility argument which specifies
  the default visibility for that mount.

- Since we want to get information about devices (for matching against
  rules) through ioctl calls to match against rules, but for userland
  to ioctl they need device nodes, dctl provides an "inner" ioctl function
  that calls the ioctl function of the device driver directly.

devfsd:
- Run in the background.

- Convert all output to use syslog(3).

- When devices are removed request the corresponding device node be
  removed from every devfs mount.

- Start using the dctl "innerioctl" facility to get information about
  devices which will be useful for matching. We're only handling disk
  devices, and can only grok information about a partition's file system
  type at the moment.

- Fix code that handles rules that specify filenames.

- Specifying visibility for device nodes now works.

mount_devfs:
- Catch up with devfs changes. Provide an -f option which means that the
  devfs can only be unmounted it if it forced.

- Provide -i option to specify that the default visibility for a mount
  should be invisible. This means device nodes are not visible in the
  devfs mount by default, rather they must have a rule for devfsd that
  specifies that they should be visible.

- Provide -v option that specifies a devfs mount should be visible.
  This is the default.

- Update man page to reflect these changes.

etc/defaults/rc.conf:
- Turn devfsd off by default.

Device nodes can now be dynamically created and removed from a device file
system.

(mjf)

- Introduce new API for allowing device drivers to register/unregister
  names for device nodes along with a corresponding dev_t.

- Make device drivers that technically never get attached and need device
  nodes (mem, zero, null) provide an initialisation function, which gets
  an entry in a table of init functions that devsw_init() calls when the
  device switch tables are initialised.

- Since we're moving to a new way of notifying devfsd(8) of new devices,
  we no longer need to link together struct devices.

(mjf)

- Special case mounting a devfs from init(8) so that a devfs cannot be
  unmounted unless it is forced. This is a (possibly temporary) workaround
  to the problem that the /etc/rc.d/root script umounts all file systems
  before mounting /.

- Clean up devfs_create_node() based on feedback from wrstuden@.

- Add devfs_remove_node() (counterpart to devfs_create_node).

- Add function to create a console and dctl(4) device node for every
  devfs mount.

- Support for VOP_MKNOD() on a devfs file system has been dropped, so
  that devfsd(8) can fully control device nodes that are created in devfs.

- Allow devfs to be mounted -o union.

(mjf)