mbuf cluster, copy the data to a separate mbuf that do not use a
cluster. this change will reduce the possiblity of packet loss
in the socket layer.
Obtained from: KAME
256 raw receive buffers (96 byte each) fit into one page. This breaks the
limit imposed by the usage of an uint8_t for the buffer number. Restrict
the allocation size for buffers to a maximum of 8192.
- Add an IPI based mechanism for migrating kses. This mechanism is
broken down into several components. This is intended to reduce cache
thrashing by eliminating most cases where one cpu touches another's
run queues.
- kseq_notify() appends a kse to a lockless singly linked list and
conditionally sends an IPI to the target processor. Right now this is
protected by sched_lock but at some point I'd like to get rid of the
global lock. This is why I used something more complicated than a
standard queue.
- kseq_assign() processes our list of kses that have been assigned to us
by other processors. This simply calls sched_add() for each item on the
list after clearing the new KEF_ASSIGNED flag. This flag is used to
indicate that we have been appeneded to the assigned queue but not
added to the run queue yet.
- In sched_add(), instead of adding a KSE to another processor's queue we
use kse_notify() so that we don't touch their queue. Also in sched_add(),
if KEF_ASSIGNED is already set return immediately. This can happen if
a thread is removed and readded so that the priority is recorded properly.
- In sched_rem() return immediately if KEF_ASSIGNED is set. All callers
immediately readd simply to adjust priorites etc.
- In sched_choose(), if we're running an IDLE task or the per cpu idle thread
set our cpumask bit in 'kseq_idle' so that other processors may know that
we are idle. Before this, make a single pass through the run queues of
other processors so that we may find work more immediately if it is
available.
- In sched_runnable(), don't scan each processor's run queue, they will IPI
us if they have work for us to do.
- In sched_add(), if we're adding a thread that can be migrated and we have
plenty of work to do, try to migrate the thread to an idle kseq.
- Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into
consideration.
- No longer use kseq_choose() to steal threads, it can lose it's last
argument.
- Create a new function runq_steal() which operates like runq_choose() but
skips threads based on some criteria. Currently it will not steal
PRI_ITHD threads. In the future this will be used for CPU binding.
- Create a kseq_steal() that checks each run queue with runq_steal(), use
kseq_steal() in the places where we used kseq_choose() to steal with
before.
the rstack functionality:
1. Fix a KASSERT that tests for the address to be above the upward
growable stack. Typically for rstack, the faulting address can be
identical to the record end of the upward growable entry, and
very likely is on ia64. The KASSERT tested for greater than, not
greater equal, so whenever the register stack had to be grown
the assertion fired.
2. When we grow the upward growable stack entry and adjust the
unlying object, don't forget to adjust the size of the VM map.
Not doing so would trigger an assert in vm_mapzdtor().
Pointy hat: marcel (for not testing with INVARIANTS).
those cylinder groups that have at least 75% of the average free
space per cylinder group for that file system are considered as
candidates for the creation of a new directory. The previous formula
for minbfree would set it to zero if the file system was more than
75% full, which allowed cylinder groups with no free space at all
to be chosen as candidates for directory creation, which resulted
in an expensive search for free blocks for each file that was
subsequently created in that directory.
Modify the calculation of minifree in the same way.
Decrease maxcontigdirs as the file system fills to decrease the
likelyhood that a cluster of directories will overflow the available
space in a cylinder group.
Reviewed by: mckusick
Tested by: kmarx@vicor.com
MFC after: 2 weeks
that libtool-using packages seem to love using this flag.
/usr/include/sys/cdefs.h:184:5: warning: "__STDC_VERSION__" is not defined
/usr/include/sys/cdefs.h:372:5: warning: "_POSIX_C_SOURCE" is not defined
/usr/include/sys/cdefs.h:378:5: warning: "_POSIX_C_SOURCE" is not defined
routine that takes a locked routing table reference and removes all
references to the entry in the various data structures. This
eliminates instances of recursive locking and also closes races
where the lock on the entry had to be dropped prior to calling
rtrequest(RTM_DELETE). This also cleans up confusion where the
caller held a reference to an entry that might have been reclaimed
(and in some cases used that reference).
Supported by: FreeBSD Foundation
are now in the header of the external buffer itself which allows us
to manipulate them in the free routine without having to lock the softc
structure or the free list. To get space for these flags the chunk number
is reduced to 8 bit which amounts to a maximum of 256 chunks per allocated
page. This restriction is now enforced by a CTASSERT.
structures come out the right size.
Fix the ones that broke. stat32 had some missing fields from the end
and statfs32 was broken due to the strange definition of MNAMELEN
(which is dependent on sizeof(long))
I'm not sure if this fixes any actual problems or not.
the module. Previously we grabbed the mutex used by the callouts,
then stopped the callout with callout_stop, but if the callout
was already active and blocked by the mutex then it would continue
later and reference the mutex after it was destroyed. Instead
stop the callout first then lock.
Supported by: FreeBSD Foundation
o add some more debugging help for figuring out why folks are
getting complaints about releasing routing table entries with
a zero refcnt
o fix comment that talked about spl's
o remove duplicate define of DUMMYNET_DEBUG
Supported by: FreeBSD Foundation
direct dispatch) to avoid extensive kernel stack usage and to
avoid directly re-entering the network stack. The latter causes
locking problems when, for example, a complete TCP handshake`
happens w/o a context switch.
clobbers this variable. Long ago, when the idle loop wasn't in a
process, it set switchtime.tv_sec to zero to indicate that the time
needs to be read after the idle loop finishes. The special case for
this isn't needed now that there is an idle process (for each CPU).
The time is read in the normal way when the idle process is switched
away from. The seconds component of the time is only zero for the
first second after the uptime is set, and the mostly-dead code was only
executed during this time. (This was slightly broken by using uptimes
instead of times relative to the Epoch -- in the original version the
seconds component of the time was only 0 for the first second after
the Epoch.)
In mi_switch(), moved the setting of switchticks to just after the
first (and now only) setting of switchtime. This setting used to be
delayed since a late setting was needed for the idle case and an early
setting was not needed. Now the early setting is needed so that
fork_exit() doesn't need to set either switchtime or switchticks.
Removed now-completely-rotted comment attached to this. Most of the
code described by the comment had already moved to sched_switch().
very first cell in the mbuf should have a cell header word (of which
everything except the payload type and the CLP bit is ignored). All
other cells should be 48 byte and get the same header as the first cell.
This fixes a problem with sending more than 120000 raw cells/sec through
an HE155. The card seems to need 2 cell times to DMA the transmit buffer
ready queue entry and the transmit buffer descriptor so at 1/3 the
link rate the transmit buffer ready queue starts to fill up. Even with this
patch it's obviously impossible to send raw cells at link rate.
- implement the tunnel egress rule in ip_ecn_egress() in ip_ecn.c.
make ip{,6}_ecn_egress() return integer to tell the caller that
this packet should be dropped.
- handle ECN at fragment reassembly in ip_input.c and frag6.c.
Obtained from: KAME
begin with sched_lock held but not recursed, so this variable was
always 0.
Removed fixup of sched_lock.mtx_recurse after context switches in
sched_switch(). Context switches always end with this variable in the
same state that it began in, so there is no need to fix it up. Only
sched_lock.mtx_lock really needs a fixup.
Replaced fixup of sched_lock.mtx_recurse in fork_exit() by an assertion
that sched_lock is owned and not recursed after it is fixed up. This
assertion much match the one in mi_switch(), and if sched_lock were
recursed then a non-null fixup of sched_lock.mtx_recurse would probably
be needed again, unlike in sched_switch(), since fork_exit() doesn't
return to its caller in the normal way.
Correct a bug when the number of pages for external mbufs was
very large. In this case the page number could overflow into the large
buffer flag. Make this more unlikley by move that flag further away.
is returned from the card to the driver. Add a counter that shows
how many times this allocation has failed. Note, that we could even
further delay the allocation of the mbuf until we know, that we need it
(there are no receive errors and the connection is open). This will be done
in a later commit.
Print the new statistics field in atmconfig.
atomic instructions instead. Remove the stuff used to track
whether an external mbuf travels through the system. This is
temporary only and will come back soon.
code has the typical branch prediction detour, which creates cross-
section branches. A LINT kernel is apparently large enough nowadays
that the .text and .text2 sections cannot always be layed-out so that
branches between them reach.
The fix is to stop using the alpha-specific bitops and instead use
the portable implementation used by all platforms other than alpha
and i386.
with an mbuf until it is reclaimed. This is in contrast to tags that
vanish when an mbuf chain passes through an interface. Persistent tags
are used, for example, by MAC labels.
Add an m_tag_delete_nonpersistent function to strip non-persistent tags
from mbufs and use it to strip such tags from packets as they pass through
the loopback interface and when turned around by icmp. This fixes problems
with "tag leakage".
Pointed out by: Jonathan Stone
Reviewed by: Robert Watson
reset in a child process after a fork(). Currently, however, only the
real timer is cleared while the virtual and profiling timers are inherited.
The realtimer is cleared because it lives directly in struct proc in
p_realtimer. It is in the zero'd section of struct proc. The other timers
live in the p_timer[] array in struct pstats. These timers are copied on
fork() rather than zero'd. The fix is to move p_timer[] to the zero'd
part of struct pstats so that they are zero'd instead of copied on fork().
Note: Since at least FreeBSD 2.0 (and possibly earlier) we've had storage
for two real interval timers. Now that the uarea is less important,
perhaps we could move all of p_timer[] over to struct proc and drop the
p_realtimer special case to fix that.
PR: kern/58647
Reported by: Dan Nelson <dnelson@allantgroup.com>
MFC after: 1 week
the RNAT bit index constant. The net effect of this is that there's
no discontinuity WRT NaT collections which greatly simplifies certain
operations. The cost of this is that there can be up to 504 bytes of
unused stack between the true base of the kernel stack and the start
of the RSE backing store. The cost of adjusting the backing store
pointer to keep the RNAT bit index constant, for each kernel entry,
is negligible.
The primary reasons for this change are:
1. Asynchronuous contexts in KSE processes have the disadvantage of
having to copy the dirty registers from the kernel stack onto the
user stack. The implementation we had so far copied the registers
one at a time without calculating NaT collection values. A process
that used speculation would not work. Now that the RNAT bit index
is constant, we can block-copy the registers from the kernel stack
to the user stack without having to worry about NaT collections.
They will be in the right place on the user stack.
2. The ndirty field in the trapframe is now also usable in userland.
This was previously not the case because ndirty also includes the
space occupied by NaT collections. The value could be off by 8,
depending on the discontinuity. Now that the RNAT bit index is
contants, we have exactly the same number of NaT collection points
on the kernel stack as we would have had on the user stack if we
didn't switch backing stores.
3. Debuggers and other applications that use ptrace(2) can now copy
the dirty registers from the kernel stack (using ptrace(2)) and
copy them whereever they want them (onto the user stack of the
inferior as might be the case for gdb) without having to worry
about NaT collections in the same way the kernel doesn't have to
worry about them.
There's a second order effect caused by the randomization of the
base of the backing store, for it depends on the number of dirty
registers the processor happened to have at the time of entry into
the kernel. The second order effect is that the RSE will have a
better cache utilization as compared to having the backing store
always aligned at page boundaries. This has not been measured and
may be in practice only minimally beneficial, if at all measurable.
license. Only clause 3 has been revoked. Restore the fourth clause
as clause 3.
Pointed out by: das@
Remove my name as a copyright holder since I don't use a BSD license
compatible or comparable to the UCB license. I choose not to add a
complete second license for my work for aesthetic reasons, nor to
replace the UCB license on grounds of rewriting more than 90% of the
source files. The rewrite can also be seen as an enhancement and since
the files were practically empty, it's rather trivial to have changed
90% of the files.
the maximum number of pages for buffers) return -1 instead of 0.
This fixes a panic under conditions when many mbufs are needed.
Update the head pointer of the receive buffer pool queue even when
we could not supply a buffer to the chip. Otherwise the chip will
not re-interrupt us for another try. A better strategy would probably
be to remember this condition and to supply buffers without an interrupt
as soon as buffers get available.
