there is an environment variable which shall initialize the SYSCTL
during early boot. This works for all SYSCTL types both statically and
dynamically created ones, except for the SYSCTL NODE type and SYSCTLs
which belong to VNETs. A new flag, CTLFLAG_NOFETCH, has been added to
be used in the case a tunable sysctl has a custom initialisation
function allowing the sysctl to still be marked as a tunable. The
kernel SYSCTL API is mostly the same, with a few exceptions for some
special operations like iterating childrens of a static/extern SYSCTL
node. This operation should probably be made into a factored out
common macro, hence some device drivers use this. The reason for
changing the SYSCTL API was the need for a SYSCTL parent OID pointer
and not only the SYSCTL parent OID list pointer in order to quickly
generate the sysctl path. The motivation behind this patch is to avoid
parameter loading cludges inside the OFED driver subsystem. Instead of
adding special code to the OFED driver subsystem to post-load tunables
into dynamically created sysctls, we generalize this in the kernel.
Other changes:
- Corrected a possibly incorrect sysctl name from "hw.cbb.intr_mask"
to "hw.pcic.intr_mask".
- Removed redundant TUNABLE statements throughout the kernel.
- Some minor code rewrites in connection to removing not needed
TUNABLE statements.
- Added a missing SYSCTL_DECL().
- Wrapped two very long lines.
- Avoid malloc()/free() inside sysctl string handling, in case it is
called to initialize a sysctl from a tunable, hence malloc()/free() is
not ready when sysctls from the sysctl dataset are registered.
- Bumped FreeBSD version to indicate SYSCTL API change.
MFC after: 2 weeks
Sponsored by: Mellanox Technologies
When safety requirements are met, it allows to avoid passing I/O requests
to GEOM g_up/g_down thread, executing them directly in the caller context.
That allows to avoid CPU bottlenecks in g_up/g_down threads, plus avoid
several context switches per I/O.
The defined now safety requirements are:
- caller should not hold any locks and should be reenterable;
- callee should not depend on GEOM dual-threaded concurency semantics;
- on the way down, if request is unmapped while callee doesn't support it,
the context should be sleepable;
- kernel thread stack usage should be below 50%.
To keep compatibility with GEOM classes not meeting above requirements
new provider and consumer flags added:
- G_CF_DIRECT_SEND -- consumer code meets caller requirements (request);
- G_CF_DIRECT_RECEIVE -- consumer code meets callee requirements (done);
- G_PF_DIRECT_SEND -- provider code meets caller requirements (done);
- G_PF_DIRECT_RECEIVE -- provider code meets callee requirements (request).
Capable GEOM class can set them, allowing direct dispatch in cases where
it is safe. If any of requirements are not met, request is queued to
g_up or g_down thread same as before.
Such GEOM classes were reviewed and updated to support direct dispatch:
CONCAT, DEV, DISK, GATE, MD, MIRROR, MULTIPATH, NOP, PART, RAID, STRIPE,
VFS, ZERO, ZFS::VDEV, ZFS::ZVOL, all classes based on g_slice KPI (LABEL,
MAP, FLASHMAP, etc).
To declare direct completion capability disk(9) KPI got new flag equivalent
to G_PF_DIRECT_SEND -- DISKFLAG_DIRECT_COMPLETION. da(4) and ada(4) disk
drivers got it set now thanks to earlier CAM locking work.
This change more then twice increases peak block storage performance on
systems with manu CPUs, together with earlier CAM locking changes reaching
more then 1 million IOPS (512 byte raw reads from 16 SATA SSDs on 4 HBAs to
256 user-level threads).
Sponsored by: iXsystems, Inc.
MFC after: 2 months
disable GEOM tasting to avoid the "bouncing GEOM" problem where, when
you shut down the consumer of a provider which can be viewed in multiple
ways (typically a mirror whose members are labeled partitions), GEOM
will immediately taste that provider's alter ego and reattach the
consumer.
Approved by: re (glebius)
and replace tsleep(9) with msleep(9) which doesn't use a timeout. The
previously used timeout caused the event process to wake up ten times
per second on an idle system.
one_event() is now called with the topology lock held and it returns
with both the topology and event locks held when there are no more
events in the queue.
Reported by: mav, Marius Nünnerich
Reviewed by: freebsd-geom
- Use thread_lock() rather than sched_lock for per-thread scheduling
sychronization.
- Use the per-process spinlock rather than the sched_lock for per-process
scheduling synchronization.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
While we wait for holds to be released, print a list of who holds us
back once per second.
Use the new KPI from GEOM instead of vfs_mount.c calling g_waitidle().
Use the new KPI also from ata.
With ATAmkIII's newbusification, ata could narrowly miss the window
and ad0 would not exist when we tried to mount root.
event posting functions varargs to fill these.
Attribute g_call_me() to appropriate g_geom's where necessary.
Add a flag argument to g_call_me() methods which will be used to signal
cancellation of events in the future.
This commit should be a no-op.
disk I/O processing.
The intent is that the disk driver in its hardware interrupt
routine will simply schedule the bio on the task queue with
a routine to finish off whatever needs done.
The g_up thread will then schedule this routine, the likely
outcome of which is a biodone() which queues the bio on
g_up's regular queue where it will be picked up and processed.
Compared to the using the regular taskqueue, this saves one
contextswitch.
Change our scheduling of the g_up and g_down queues to be water-tight,
at the cost of breaking the userland regression test-shims.
Input and ideas from: scottl
idle time.
Statistics now default to "on" and can be turned off with
sysctl kern.geom.collectstats=0
Performance impact of statistics collection is on the order of
800 nsec per consumer/provider set on a 700MHz Athlon.
Insted of embedding a struct g_stat in consumers and providers, merely
include a pointer.
Remove a couple of <sys/time.h> includes now unneeded.
Add a special allocator for struct g_stat. This allocator will allocate
entire pages and hand out g_stat functions from there. The "id" field
indicates free/used status.
Add "/dev/geom.stats" device driver whic exports the pages from the
allocator to userland with mmap(2) in read-only mode.
This mmap(2) interface should be considered a non-public interface and
the functions in libgeom (not yet committed) should be used to access
the statistics data.
Add debug.sizeof.g_stat sysctl.
Set the id field of the g_stat when we create consumers and providers.
Remove biocount from consumer, we will use the counters in the g_stat
structure instead. Replace one field which will need to be atomically
manipulated with two fields which will not (stat.nop and stat.nend).
Change add companion field to bio_children: bio_inbed for the exact
same reason.
Don't output the biocount in the confdot output.
Fix KASSERT in g_io_request().
Add sysctl kern.geom.collectstats defaulting to off.
Collect the following raw statistics conditioned on this sysctl:
for each consumer and provider {
total number of operations started.
total number of operations completed.
time last operation completed.
sum of idle-time.
for each of BIO_READ, BIO_WRITE and BIO_DELETE {
number of operations completed.
number of bytes completed.
number of ENOMEM errors.
number of other errors.
sum of transaction time.
}
}
API for getting hold of these statistics data not included yet.
WARNING: This is not a published interface, it is a stopgap measure for
WARNING: libdisk so we can get 5.0-R out of the door.
Sponsored by: DARPA & NAI Labs
This is not quite the set of information I would want, but the tree where
I have the "correct" version is messed up with conflicts.
Sponsored by: DARPA & NAI Labs.
work.
This prevents people from sleeping in the UP/DOWN I/O path by mistake
or design (doing so almost invariably result in deadlocks since it
stalls all I/O processing in the given direction.
Sponsored by: DARPA & NAI Labs.
Add a __unused.
Make the 2byte decoder functions return 16 bits for the benefits
of picky lints.
No need to grab giant around a tsleep() when we have a timeout.
Sponsored by: DARPA & NAI Labs.
test and play with this.
This is not yet production quality and should be run only on dedicated
test boxes.
For people who want to develop transformations for GEOM there exist a
set of shims to run geom in userland (ask phk@freebsd.org).
Reports of all kinds to: phk@freebsd.org
Please include in report:
dmesg
sysctl debug.geomdot
sysctl debug.geomconf
Known significant limitations:
no kernel dump facility.
ioctls severely restricted.
Sponsored by: DARPA, NAI Labs
