- Remove the double queue mechanism for timeshare threads. It was slow
due to excess cache lines in play, caused suboptimal scheduling behavior
with niced and other non-interactive processes, complicated priority
lending, etc.
- Use a circular queue with a floating starting index for timeshare threads.
Enforces fairness by moving the insertion point closer to threads with
worse priorities over time.
- Give interactive timeshare threads real-time user-space priorities and
place them on the realtime/ithd queue.
- Select non-interactive timeshare thread priorities based on their cpu
utilization over the last 10 seconds combined with the nice value. This
gives us more sane priorities and behavior in a loaded system as
compared to the old method of using the interactivity score. The
interactive score quickly hit a ceiling if threads were non-interactive
and penalized new hog threads.
- Use one slice size for all threads. The slice is not currently
dynamically set to adjust scheduling behavior of different threads.
- Add some new sysctls for scheduling parameters.
Bug fixes/Clean up:
- Fix zeroing of td_sched after initialization in sched_fork_thread() caused
by recent ksegrp removal.
- Fix KSE interactivity issues related to frequent forking and exiting of
kse threads. We simply disable the penalty for thread creation and exit
for kse threads.
- Cleanup the cpu estimator by using tickincr here as well. Keep ticks and
ltick/ftick in the same frequency. Previously ticks were stathz and
others were hz.
- Lots of new and updated comments.
- Many many others.
Tested on: up x86/amd64, 8way amd64.
- runq_add_pri allows the caller to position the thread at any rqindex
regardless of priority.
- runq_choose_from() chooses the lowest priority thread starting from a given
index. The index is updated with the rqindex of the chosen thread. This
routine is used to pick the lowest priority relative to a given index.
- runq_remove_idx() updates the index if the run queue that held the removed
thread is now empty.
start working with third party usb modules, where sometimes it
is not easy to set the inclusion order so that there are no multiple
inclusions, yet you want to compile with high WARNS levels).
I am not sure if there is a standard for having a leading and/or trailing _
in the macro name, the usb code seems to use both.
There are still several unprotected headers here so it might be useful
to do the same thing on other files as well as the need arises.
MFC After: 3 days
check length of the pathname in the range 0<=n<=NFS_MAXPATHLEN,
not 0<n<=NFS_MAXPATHLEN. This fixes a minor interoperability problem
that the FreeBSD NFS server did not allow a symlink pointing the empty
pathname.
MFC after: 1 week
mbuf. First moves toward being able to cope better with having layer 2 (or
other encapsulation data) before the IP header in the packet being examined.
More commits to come to round out this functionality. This commit should
have no practical effect but clears the way for what is coming.
Revirewed by: luigi, yar
MFC After: 2 weeks
been introduced to the MAC framework:
mpo_associate_nfsd_label
mpo_create_mbuf_from_firewall
mpo_check_system_nfsd
mpo_check_vnode_mmap_downgrade
mpo_check_vnode_mprotect
mpo_init_syncache_label
mpo_destroy_syncache_label
mpo_init_syncache_from_inpcb
mpo_create_mbuf_from_syncache
MFC after: 2 weeks [1]
[1] The syncache related entry points will NOT be MFCed as the changes in
the syncache subsystem are not present in RELENG_6 yet.
exclusive access if there is at least one thread waiting for it to
become available. This may significantly reduce overhead by reducing
the number of unnecessary wakeups issued whenever the framework becomes
idle.
Annotate that we still signal the CV more than necessary and should
fix this.
Obtained from: TrustedBSD Project
Reviewed by: csjp
Tested by: csjp
Redo the checking for 2.6 emulation. We now cache the value of
use26 and replace calls to linux_get_osrelease() + parsing with
a call to linux_use26(). Typical path is lockless now.
Pointed out by: kib
This allows to ship RELENG_7_0 with a default osrelease of 2.4.2 and the
possibility to enable 2.6.x emulation without the possible performance
impact of the previous version of the check.
Submitted by: rdivacky
- Micro-optimize the addition of an 802.1q header to match the removal code.
- Consistently check for interfaces being up and running.
- Consistently use NULL instead of 0 with pointers.
With the second (and last) part of my previous Summer of Code work, we get:
-ipfw's in kernel nat
-redirect_* and LSNAT support
General information about nat syntax and some examples are available
in the ipfw (8) man page. The redirect and LSNAT syntax are identical
to natd, so please refer to natd (8) man page.
To enable in kernel nat in rc.conf, two options were added:
o firewall_nat_enable: equivalent to natd_enable
o firewall_nat_interface: equivalent to natd_interface
Remember to set net.inet.ip.fw.one_pass to 0, if you want the packet
to continue being checked by the firewall ruleset after being
(de)aliased.
NOTA BENE: due to some problems with libalias architecture, in kernel
nat won't work with TSO enabled nic, thus you have to disable TSO via
ifconfig (ifconfig foo0 -tso).
Approved by: glebius (mentor)
access plus timers. This makes the code
more portable and able to change out the
mbuf or timer system used more easily ;-)
b) removal of all use of pkt-hdr's until only
the places we need them (before ip_output routines).
c) remove a bunch of code not needed due to <b> aka
worrying about pkthdr's :-)
d) There was one last reorder problem it looks where
if a restart occur's and we release and relock (at
the point where we setup our alias vtag) we would
end up possibly getting the wrong TSN in place. The
code that fixed the TSN's just needed to be shifted
around BEFORE the release of the lock.. also code that
set the state (since this also could contribute).
Approved by: gnn
semantics.
- Stop testing bpf pointers for NULL. In some cases use
bpf_peers_present() and then call the function directly inside the
conditional block instead of the macro.
- For places where the entire conditional block is the macro, remove the
test and make the macro unconditional.
- Use BPF_MTAP() in if_pfsync on FreeBSD instead of an expanded version of
the old semantics.
Reviewed by: csjp (older version)
lookup early. This has some performance implications and should not be
enabled by default, but might help greatly in certain setups. After some
more testing this could be turned into a sysctl.
Tested by: avatar
LOR ids: 17, 24, 32, 46, 191 (conceptual)
MFC after: 6 weeks
MPLOCKED. The cleaning in rev.1.25 was supposed to have been undone
by rev.1.26, but 1.26 could never have actually affected asm files
since atomic.h is full of C declarations so including it in asm files
would just give syntax errors. The asm MPLOCKED is even less needed
than when misplaced definitions of it were first removed, and is now
unused in any asm file in the src tree except in anachronismns in
sys/i386/i386/support.s.
manipulation is visible to the subject process. Remove XXX comments
suggesting this.
Convert one XXX on a difference from Darwin into a note: it's not a
bug, it's a feature.
Obtained from: TrustedBSD Project
system calls on the amd64 architecture.
Some minor white space tweaks for consistency with other syscalls.master
files.
Obtained from: TrustedBSD Project
- Replace XXX with Note: in several cases where observations are made about
future functionality rather than problems or bugs.
- Remove an XXX comment about byte order and au_to_ip() -- IP headers must
be submitted in network byte order. Add a comment to this effect.
- Mention that we don't implement select/poll for /dev/audit.
Obtained from: TrustedBSD Project
kernel<->policy ABI version. Add a comment to the definition describing
it and listing known versions. Modify MAC_POLICY_SET() to reference the
current kernel version by name rather than by number.
Staticize mac_late, which is used only in mac_framework.c.
Obtained from: TrustedBSD Project
mac_framework.c Contains basic MAC Framework functions, policy
registration, sysinits, etc.
mac_syscalls.c Contains implementations of various MAC system calls,
including ENOSYS stubs when compiling without options
MAC.
Obtained from: TrustedBSD Project
consumes and implements, as well as the location of the framework and
policy modules.
Refactor MAC Framework versioning a bit so that the current ABI version can
be exported via a read-only sysctl.
Further update comments relating to locking/synchronization.
Update copyright to take into account these and other recent changes.
Obtained from: TrustedBSD Project
node would send every outgoing frame to the "compress" hook.
Packets received on the "compress" hook were expected to be
compressed and PROT_COMPD tag was put on them unconditionally.
After this commit an alternative compression mode can be set.
In this mode the node doesn't put the PROT_COMPD, the compressor
should put it itself. This is important for such kind of
compressors, that can submit uncompressed frames.
Before this commit, if the decompression is enabled, the ng_ppp(4)
node would send and incoming frame to the "decompress" hook
only if it has the PROT_COMPD proto tag on it.
After this commit an alternative decompression mode can be set.
In this mode the node sends all the incoming packets to the
decompression hook. This is important for such kind of compressors
that need uncompressed packets too, to keep their library in sync.
These new features will be used in new version of mpd4, and in new
compressor nodes.
Submitted by: Alexander Motin <mav alkar.net>
mainly involves removing all __CC_SUPPORTS___INLINE__ ifdefs. These
ifdefs are even less needed for amd64 than for i386, but the i386
atomic.h never had them. The ifdefs here were just an optimization
of obsolescent compatibility cruft (__inline) for a null set of
compilers. I think null sets of compilers should only be supported
in cases where this is more than an optimization, doesn't require
extensive ifdefs, and only involves not-so-obsolescent compatibility
cruft (plain inline here).
o mark 11g mode support on finding 11g or pure 11g (OFDM-only)
channels; was requiring pure 11g which caused some contortions
in drivers that manually setup their channel lists
These functions are used a lot for mutexes, so this reduces the text
size of an average kernel by about 0.75%. This wasn't intended to
be a significant optimization, but it somehow increased the maximum
number of packets per second that can be transmitted by my bge hardware
from 320000 to 460000 (this benchmark is CPU-bound and remarkably
sensitive to changes in the text section).
Details: we would prefer to leave the result of the cmpxchg in %al,
but cannot tell gcc that it is there, so we have to convert it to an
integer register. We converted to %al, then to %[re]ax, but the
latter step is usually wasted since gcc usually only wants the condition
code and can recover it from %al just as easily as from %[re]ax. Let
gcc promote %al in the few cases where this is needed.
Nearby style fixes;
- let gcc manage the load of `res', and don't abuse `res' for a copy of `exp'
- don't echo `res's name in comments
- consistently spell the condition code as 'e' after comparison for equality
- don't hard-code %al anywhere except in constraints
- for the version that doesn't use cmpxchg, there is no requirement to use
%al anywhere, so don't hard-code it in the constraints either.
Style non-fix:
- for the versions that use cmpxchg, keep using "a" (was %[re]ax, now %al)
for the main output operand, although this is not required. The input
and output operands that use the "a" constraint are now decoupled, and
this makes things clearer except for the reason that the output register
is hard-coded. It is now just a hack to tell gcc that the input "a" has
been clobbered without increasing the number of operands.
o change handling of regdomain-related mib knobs so they can be set
post-attach: regdomain, countrycode, outdoor, and xchanmode; the
hal will not permit changing the regdomain but we expose it for now
o on regdomain/countrycode change recalculate the channel list and
push it to the net80211 layer (NB: looks to need more tweaking)
o setup rate tables for half/quarter rate channels
o honor half/quarter rate channel configs when changing channels
o honor half/quarter rate channel configs when setting the slot time
o use hack/nonstandard channel numbering scheme for the public safety
band to avoid overlapping 2.4G channels on dual-band cards
o remove setup of ic_sup_rates; the net80211 layer can do this for us
and it simplifies handling of half/quarter rate channels
Tested only in Public Safety Band with cards that have RF5112.
in the Public Safety Band):
o add channel flags to identify half/quarter-rate operation
o add rate sets (need to check spec on 4Mb/s in 1/4 rate)
o add if_media definitions for new rates
o split net80211 channel setup out into ieee80211_chan_init
o fixup ieee80211_mhz2ieee and ieee80211_ieee2mhz to understand half/quarter
rate channels: note we temporarily use a nonstandard/hack numbering that
avoids overlap with 2.4G channels because we don't (yet) have enough
state to identify and/or map overlapping channel sets
o fixup ieee80211_ifmedia_init so it can be called post attach and will
recalculate the channel list and associated state; this enables changing
channel-related state like the regulatory domain after attach (will be
needed for 802.11d support too)
o add ieee80211_get_suprates to return a reference to the supported rate
set for a given channel
o add 3, 4.5, and 27 MB/s tx rates to rate <-> media conversion routines
o const-poison channel arg to ieee80211_chan2mode
bge_intr(). Some of them are used in bge_poll(). Simplify by only
initializing these for polling mode and not toggling them when switching
modes. This also fixes missing synchronization with the coalescing
engine in the toggling.
Add a pointer to the relevant PR for future reference. The whole comment
will be OK to remove as soon as the general solution is applied.
PR: kern/105943
pmap on i386
- check for change in executable status in pmap_enter
- pmap_qenter and pmap_qremove only need to invalidate the range if one
of the pages has been referenced
- remove pmap_kenter/pmap_kremove as they were only used by pmap_qenter
and pmap_qremove
- in pmap_copy don't copy wired bit to destination pmap
- mpte was unused in pmap_enter_object - remove
- pmap_enter_quick_locked is not called on the kernel_pmap, remove check
- move pmap_remove_write specific logic out of tte_clear_phys_bit
- in pmap_protect check for removal of execute bit
- panic in the presence of a wired page in pmap_remove_all
- pmap_zero_range can call hwblkclr if offset is zero and size is PAGE_SIZE
- tte_clear_virt_bit is only used by pmap_change_wiring - thus it can be
greatly simplified
- pmap_invalidate_page need only be called in tte_clear_phys_bit if there
is a match with flags
- lock the pmap in tte_clear_phys_bit so that clearing the page bits is
atomic with invalidating the page
- these changes result in 100s reduction in buildworld from a malloc backed
disk to a malloc backed disk - ~2.5%
mbuf is dropped, to preserve the invariant in the PR_ADDR case.
Add a regression test to detect this condition, but do not hook it
up to the build for now.
PR: kern/38495
Submitted by: James Juran
Reviewed by: sam, rwatson
Obtained from: NetBSD
MFC after: 2 weeks
The problem was that I was acquiring the driver sx lock and then waiting
for a taskqueue to drain, however the taskqueue itself would try to
acquire the lock as well leading to a deadlock.
To fix the problem roll my own exclusive lock that allows for lock
cancellation. This is a normal exclusive lock, however if someone
marks it as "dead" then all waiters who request an error return will
get back an error instead of continuing to wait for the lock.
In this particular case, the shutdown and detach functions kill the
lock while the async task thread tries to acquire the lock but will
abort if the lock returns an error.
The other option was to drop the driver lock mid-detach and mid-shutdown,
mid-detach was a ok, however mid-shutdown was not.
While I'm here, fix a bug in what appears to be the mii link status
word in the softc going out to lunch. Explicitly set the status
word to 1 after initializing the mii. This would result in an interface
that would never respond to "if_start" requests as the mii interface
would always look down.
nmi handler is used to stop other processors, nmi hander calls trap(),
however, trap() now accepts a pointer rather than a reference, this was
changed by kmacy@.
non-extattr functions from vfs_extattr.c, and extattr functions from
vfs_syscalls.c.
Change copyright/license on vfs_extattr.c to my copyright/license on
the extended attribute implementation (from extattr.h).
Clean up includes a bit.
Obtained from: TrustedBSD Project
Framework and security modules, to src/sys/security/mac/mac_policy.h,
completing the removal of kernel-only MAC Framework include files from
src/sys/sys. Update the MAC Framework and MAC policy modules. Delete
the old mac_policy.h.
Third party policy modules will need similar updating.
Obtained from: TrustedBSD Project
return an error since it returns a count of battery devices in the system.
Set it to 0 explicitly, since it is the only switch branch that doesn't set
it.
# I guess no one uses it.
It always called MH_ALIGN for small lengths being
prepended (less than MHLEN). This meant that if you did
a prepend on a non M_PKTHDR the system would panic with
the KASSERT in MH_ALIGN. Instead we are not aware of
this and do a MH_ALIGN or M_ALIGN as appropriate.
Reviewed by: andre
Approved by: gnn
subsystems will be a property of policy modules, which may require
access control check entry points to be invoked even when not actively
enforcing (i.e., to track information flow without providing
protection).
Obtained from: TrustedBSD Project
Suggested by: Christopher dot Vance at sparta dot com
than from the slab, but don't.
Document mac_mbuf_to_label(), mac_copy_mbuf_tag().
Clean up white space/wrapping for other comments.
Obtained from: TrustedBSD Project
Exapnd comments on System V IPC labeling methods, which could use improved
consistency with respect to other object types.
Obtained from: TrustedBSD Project
the ifnet itself. The stack copy has been made while holding the mutex
protecting ifnet labels, so copying from the ifnet copy could result in
an inconsistent version being copied out.
Reported by: Todd.Miller@sparta.com
Obtained from: TrustedBSD Project
MFC after: 3 weeks
- Move linux_nanosleep() from src/sys/amd64/linux32/linux32_machdep.c to
src/sys/compat/linux/linux_time.c.
- Validate timespec ranges before use as Linux kernel does.
- Fix l_timespec structure.
- Clean up style(9) nits.
Add rudimentary IPC_INFO/MSG_INFO command support for linux_msgctl()
to pacify Linux ipcs(1). While I am here, add more bound checks
for linux_msgsnd() and linux_msgrcv().
copyin()/copyout() for message type is separated from msgsnd()/msgrcv() and
it is done from its wrapper functions to support 32-bit emulations. After I
implemented this, I have briefly referenced NetBSD and Darwin. NetBSD passes
copyin()/copyout() function pointers from wrappers. Darwin passes size of
message type as an argument, which is actually similar to my first
implementation (P4 109706). We may revisit these implementations later.
would be able to work with aac(4).
This approach is used by some other drivers as well. However, we
need a more generic way to do this in order to avoid having to
special case headers in individual drivers for each platform.
Obtained from: Adaptec (version b11518)
Approved by: scottl
been handled instead of when at least one descriptor was just handled.
For bge, it is normal to get a txeof when only a small fraction of the
queued tx descriptors have been handled, so the bug broke the watchdog
in a usual case.
- moved the synchronizing bus read to after the bus write for the first
interrupt ack so that it actually synchronizes everything necessary.
We were acking not only the status update that triggered the interrupt
together with any status updates that occurred before we got around
to the bus write for the ack, but also any status updates that occur
after we do the bus write but before the write reaches the device.
The corresponding race for the second interrupt ack resulted in
sometimes returning from the interrupt handler with acked but
unserviced interrupt events. Such events then remain unserviced
until further events cause another interrupt or the watchdog times
out.
The race was often lost on my 5705, apparently since my 5705 has broken
event coalescing which causes a status update for almost every packet,
so another status update is quite likely to occur while the interrupt
handler is running. Watchdog timeouts weren't very noticeable,
apparently because bge_txeof() has one of the usual bugs resetting the
watchdog.
- don't disable device interrupts while bge_intr() is running. Doing this
just had the side effects of:
- entering a device mode in which different coalescing parameters apply.
Different coalescing parameters can be used to either inhibit or
enhance the chance of getting another status update while in the
interrupt handler. This feature is useless with the current
organization of the interrupt handler but might be useful with a
taskqueue handler.
- giving a race for ack+reenable/return. This cannot be handled
by simply rearranging the order of bus accesses like the race for
ack+keepenable/entry. It is necessary to sync the ack and then
check for new events.
- taking longer, especially with the extra code to avoid the race on
ack+reenable/return.
Reviewed by: ru, gleb, scottl
vnode v_flag. For cluster buffers this would result in dereferencing NULL
b_vp. To prevent the panic, cache relevant vnode flag before calling
bstrategy.
Reported by: Peter Holm, kris
Tested by: Peter Holm
Reviewed by: tegge
Pointy hat to: kib
running thread's id on each cpu. This allow us to add in-kernel adaptive
spin for user level mutex. While spinning in user space is possible,
without correct thread running state exported from kernel, it hardly
can be implemented efficiently without wasting cpu cycles, however
exporting thread running state unlikely will be implemented soon as
it has to design and stablize interfaces. This implementation is
transparent to user space, it can be disabled dynamically. With this
change, mutex ping-pong program's performance is improved massively on
SMP machine. performance of mysql super-smack select benchmark is increased
about 7% on Intel dual dual-core2 Xeon machine, it indicates on systems
which have bunch of cpus and system-call overhead is low (athlon64, opteron,
and core-2 are known to be fast), the adaptive spin does help performance.
Added sysctls:
kern.threads.umtx_dflt_spins
if the sysctl value is non-zero, a zero umutex.m_spincount will
cause the sysctl value to be used a spin cycle count.
kern.threads.umtx_max_spins
the sysctl sets upper limit of spin cycle count.
Tested on: Athlon64 X2 3800+, Dual Xeon 5130
re_watchdog() in order to avoid races accessing if_timer.
- Use bus_get_dma_tag() so re(4) works on platforms requiring it.
- Remove invalid BUS_DMA_ALLOCNOW when creating the parent DMA tag
and the tags that are used for static memory allocations.
- Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that.
- Remove an unused variable in re_intr().
watchdog timer in dc_txeof() in case there are still unhandled
descriptors as dc_poll() invokes dc_poll() unconditionally.
Otherwise this would result in the watchdog timer constantly being
being reloaded and thus circumvent that the watchdog ever fires in
the DEVICE_POLLING case.
Pointed out by: bde
pmap.c, and is potentially the cause of hangs reported on machines with a
small amount of memory. On machines with sufficient RAM, and without a lot
of processes running, this situation would probably never occur.
Testing is still incomplete, but it is obviously wrong so remove the
offending code now.
The issue of what to do when both the primary and secondary hash overflow
is still open.
Reported by: Dan Kresja at windriver dot com, via alc
This macro was written expecting a 32-bit unsigned long, and
doesn't work properly on 64-bit systems. This bug caused vn_stat()
to return incorrect values for files larger than 2gb on msdosfs filesystems
on 64-bit systems.
PR: 106703
Submitted by: Axel Gonzalez <loox e-shell net>
MFC after: 3 days
This bug caused vn_stat() to fail on files larger than 2gb on msdosfs
filesystems on AMD64.
PR: 106703
Tested by: Axel Gonzalez <loox e-shell net>
MFC after: 3 days
- Do not repeatedly read vendor/device IDs while probing.
- Remove redundant bzero(3) for softc. device_get_softc(9) does it for free[1].
Reviewed by: glebius
Suggested by: glebius[1]
aches as a read-only file. In a number of cases this has led to
compiles failing- usually due to some strange NFS drift which thinks
that the opt_ah.h in the compile directory is out of date wrt the
source it is copied from. When the copy is executed again, it fails
because the target is read-only. Oops. Modify the compile hooks
avoid this.
Discussed with a while back with: Sam Leffler
- If we want mii_phy_add_media() to add 1000baseT media, we need to
supply sc->mii_extcapabilities.
- Fix formatting when announcing autonegotiation support.
usage to conform to that of tl0_trap - the separate code path
for unaligned faults was never getting used (and evidently doesn't
work), so ifdef out for now
Because accessing ID registers in rtl81x9 needs 32bit register access
and RL_IDR4/RL_IDR5 registers are reservered registers bzero() is
needed before copying ethernet address.
This fixes unaligned memory accesses panic in sparc64.
PR: kern/106801
MFC after: 3 days
as if they were really passed by reference. Specifically, the dead stores
elimination pass in the GCC 4.1 optimiser breaks the non-compliant behavior
on which FreeBSD relied. This change brings FreeBSD up to date by switching
trap frames to being explicitly passed by reference.
Reviewed by: kan
Tested by: kan
passed by value (trap frames) as if they were in fact being passed by
reference. For better or worse, this incorrect behaviour is no longer
present in gcc 4.1. In this patch I convert all trapframe arguments to
be explicitly pass by reference. I also remove vm86_initflags, pushing
the very little work that it actually does up into vm86_prepcall.
Reviewed by: kan
Tested by: kan
