Provide multiple clean queues partitioned into 'domains'. Each domain manages
its own bufspace and has its own bufspace daemon. Each domain has a set of
subqueues indexed by the current cpuid to reduce lock contention on the cleanq.
Refine the sleep/wakeup around the bufspace daemon to use atomics as much as
possible.
Add a B_REUSE flag that is used to requeue bufs during the scan to approximate
LRU rather than locking the queue on every use of a frequently accessed buf.
Implement bufspace_reserve with only atomic_fetchadd to avoid loop restarts.
Reviewed by: markj
Tested by: pho
Sponsored by: Netflix, Dell/EMC Isilon
Differential Revision: https://reviews.freebsd.org/D14274
Building the swapped welcome menu (first two items swapped) is kind of a
sluggish, because it requires a full (recrusive) shallow copy of the welcome
menu. Cache the result of that and re-use it later, instead of building it
everytime.
While here, don't create temporary locals just for swapping. The following
is just as good:
x, y = y, x;
Reported by: Alexander Nasonov <alnsn@yandex.ru> (swapping)
A subtle logic bug, probably introduced in r311895, caused tail to print the
first two lines of piped input in forward order, if the very first character
was a newline.
PR: 222671
Reported by: Jim Long <freebsd-bugzilla@umpquanet.com>, pprocacci@gmail.com
MFC after: 3 weeks
Sponsored by: Spectra Logic Corp
VirtIO V1 provides configuration in multiple VENDOR capabilities so this
allows all of the configuration to be discovered.
Reviewed by: jhb
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D14325
[Enter] should be moved to the single user menu item when we swap them.
Define a non-standard menu entry function "alternate_name" to use for this
purpose for ultimate flexibility if we change our minds later. When we're
booting single user, make a shallow copy of the menu that we'd normally
display and swap the items and their name functions to use alternate_name
instead. Toggling single user in the options menu and going back to the main
menu will now correctly reflect the current boot setting with the first two
menu options and "[Enter]" will always be on the right one.
This shallow copy technique has the chance of being quite slow since it's
done on every redraw, but in my testing it does not seem to make any obvious
difference.
shallowCopyTable could likely belong better in a general-purpose utility
module, but this (and the key constnats) are the only candidates we have at
the moment so we'll drop it into our core stuff for the moment and consider
re-organization at a later date.
lualoader does a pretty good job of reverting any environment changes now.
It will even wipe out boot_verbose if it's set explicitly in loader.conf(5)
and overwritten in the boot options menu.
Future work will likely change this, as explicit choices made in the menu
should probably override the new loader.conf(5). I don't suspect this will
cause much grief, though, so it is not a high priority until boot
environment support actually lands.
This will be used when boot environment support lands to make a good-faith
effort to apply any new loader.conf(5) environment settings atop the default
configuration that we started with.
If we've fetched menu.entries and it turns out it's a function, call it to
get the actual menu entries.
This will be used to swap multi-/single- user boot options if we're booting
single user by default (boot_single="YES" in loader.conf(5)). It can also be
used fairly easily for other non-standard situations.
Instead of directly listing them in menu.welcome and menu.boot_options,
store them at menu.welcome.entries and welcome.boot_options.entries.
This will come into play later when we need to re-order the welcome menu if
boot_single is specified.
Menus are actually defined as entries in the 'menu' table. These local
declarations have not been used in the history of our in-tree lua scripts,
so give them the boot.
zero, matching the IEEE 1275 standard. Since these internal error paths
have never, to my knowledge, been taken, behavior is unchanged.
Reported by: gonzo
MFC after: 2 weeks
Loading the kernel and modules can be really slow. Loading before the menu
draws and every time one changes kernel/boot environment is even more
painful.
Defer loading until we either boot, auto-boot, or escape to loader prompt.
We still need to deal with configuration changes as the boot environment
changes, but this is generally much quicker.
This commit strips all ELF loading out of config.load/config.reload so that
these are purely for configuration. config.loadelf has been created to deal
with kernel/module loads. Unloading logic has been ripped out, as we won't
need to deal with it in the menu anymore.
Discussed in part with: allanjude
Since r323578 we may remove the last reference to a covered vnode with
vrele() instead of vput(). So, v_usecount may be decremented before
the vnode is locked and zfsctl_snapdir_lookup may "catch" the vnode
with v_usecount of zero and v_holdcnt of one.
PR: 225795
Reported by: asomers
MFC after: 1 week
r329550 introduced config.kernel_loaded. config.load() doesn't provide a
means of overriding the kernel to load, but that likely isn't necessary as
it will not be a common case. When loading the kernel, just attempt to load
the kernel previously loaded and specified in config.kernel_loaded.
If we haven't loaded a kernel yet, config.kernel_loaded will be unset/nil
and the "default"/first kernel found will be loaded. If we've loaded a
kernel, we'll try to load that same kernel again and fallback to the default
kernel if we need to.
This in also in support of upcoming boot environment support.
'nil' means the 'first kernel found in module_path', which is the same
interpretation as passing 'nil' to loadkernel.
Otherwise, it denotes the name of a kernel that we've successfully loaded.
When reloaded later, we will still need to do the full search again to
locate the actual kernel in case things have changed, so just the name is
good enough.
This is in support of upcoming boot environment support. vfs.root.mountfrom
and currdev will be changed, then we will reload configuration and attempt
to reload the currently chosen kernel unless we shouldn't.
In the worst case scenario, we have no passwords to prompt for and we end up
just clearing the screen twice before we draw the menu or proceed with boot.
In the best case scenario, we don't try drawing password prompts amidst a
bunch of kernel/module loading.
This covers the lua style guidelines we've generally agreed on so far. It
will be revised as work continues and we run into more scenarios that need
specified.
Discussed with: cem, jilles
Differential Revision: https://reviews.freebsd.org/D14423
The race manifested itself mostly in terms of crashes with "spin lock
held too long".
Relevant parts of respective code paths:
exit: reap:
PROC_LOCK(p);
PROC_SLOCK(p);
p->p_state == PRS_ZOMBIE
PROC_UNLOCK(p);
PROC_LOCK(p);
/* exit work */
if (p->p_state == PRS_ZOMBIE) /* true */
proc_reap()
free proc
/* more exit work */
PROC_SUNLOCK(p);
Thus a still exiting process is reaped.
Prior to the change the zombie check was followed by slock/sunlock trip
which prevented the problem.
Even code prior to this commit has a bug: the proc is still accessed for
statistic collection purposes. However, the severity is rather small and
the bug may be fixed in a future commit.
Reported by: many
Tested by: allanjude
The primitive can be used to wait for the lock to be released. Intended
usage is for locks in structures which are about to be freed.
The benefit is the avoided interrupt enable/disable trip + atomic op to
grab the lock and shorter wait if the lock is held (since there is no
worry someone will contend on the lock, re-reads can be more aggressive).
Briefly discussed with: kib
Despite best efforts to regularize, there's a few tables in the system
that still report they are for bus usb when they are really for bus
uhub (where usb devices attach). Add a temporary workaround for this
until these places have been eliminated (likely my fault).
Second, when running verbose, describe what we're doing when
searching. This output can be quite long, but says exactly what's
going on (this output is to stdout, so it's useless for scripting).
and index. A private function to do exactly that already existed, so this
renames gpio_pin_get_by_ofw_impl() to gpio_pin_get_by_ofw_propidx() and
provides a declaration for it in a public header.
Previously there were functions to get a pin by property name (assuming
there would only be one pin defined for the name), or by index (asuming
the property has the standard name "gpios"). It turns out there are
devicetree bindings that describe properties with names other than "gpios"
which can describe multiple pins. Hence the need to retrieve the Nth item
from a named property.
