o sort mbuf flags together and extend values to 32 bits
o write M_COPYFLAGS in terms of M_PROTOFLAGS
o move M_COPYFLAGS and M_PROTOFLAGS up to be together with flag defs
Reviewed by: rwatson
MFC after: 3 weeks
- Take advantage of m_collapse(9).
- Sync with other NIC drivers and prepend a TX mbuf if the first attempt
to load it fails with an error other than EFBIG and stop trying instead
of freeing it and keeping on trying to enqueue more mbufs. Also ensure
the driver queue isn't empty before trying to enqueue mbufs in order to
reduce locking operations.
- In xl_ifmedia_upd() add a missing XL_UNLOCK(). [1]
- Const'ify the xl_devs array.
- Remove an outdated comment.
PR: 113406 [1]
MFC after: 1 month
- Correct the maxsize parameter when creating the mbufs busdma tag to
reflect the actual requirement of dc(4).
- Move the KASSERT in dc_newbuf() to the right spot.
- Also convert the TX side to take advantage of bus_dmamap_load_mbuf_sg(9).
- Move the comment regarding dc_start_locked() to the right spot.
MFC after: 2 weeks
- Resource allocation in aac_alloc (moved from from aac_init)
- Interrupt setup in aac_setup_intr (from aac_attach)
- Container probing in aac_get_container_info (from aac_startup and
aac_handle_aif)
- Firmware status check moved to aac_check_firmware from aac_init
In case of "new SA", we must check the hard lifetime of the old SA
to find out if it is not permanent and we can delete it.
Submitted by: sakane via gnn
MFC after: 3 days
overhead of packet capture by allowing a user process to directly "loan"
buffer memory to the kernel rather than using read(2) to explicitly copy
data from kernel address space.
The user process will issue new BPF ioctls to set the shared memory
buffer mode and provide pointers to buffers and their size. The kernel
then wires and maps the pages into kernel address space using sf_buf(9),
which on supporting architectures will use the direct map region. The
current "buffered" access mode remains the default, and support for
zero-copy buffers must, for the time being, be explicitly enabled using
a sysctl for the kernel to accept requests to use it.
The kernel and user process synchronize use of the buffers with atomic
operations, avoiding the need for system calls under load; the user
process may use select()/poll()/kqueue() to manage blocking while
waiting for network data if the user process is able to consume data
faster than the kernel generates it. Patchs to libpcap are available
to allow libpcap applications to transparently take advantage of this
support. Detailed information on the new API may be found in bpf(4),
including specific atomic operations and memory barriers required to
synchronize buffer use safely.
These changes modify the base BPF implementation to (roughly) abstrac
the current buffer model, allowing the new shared memory model to be
added, and add new monitoring statistics for netstat to print. The
implementation, with the exception of some monitoring hanges that break
the netstat monitoring ABI for BPF, will be MFC'd.
Zerocopy bpf buffers are still considered experimental are disabled
by default. To experiment with this new facility, adjust the
net.bpf.zerocopy_enable sysctl variable to 1.
Changes to libpcap will be made available as a patch for the time being,
and further refinements to the implementation are expected.
Sponsored by: Seccuris Inc.
In collaboration with: rwatson
Tested by: pwood, gallatin
MFC after: 4 months [1]
[1] Certain portions will probably not be MFCed, specifically things
that can break the monitoring ABI.
references to a vnode with VI_OWEINACT set will force the vinactive()
call. The kernel makes no guarantees about which reference was the
last to close a file or when the actual inactive processing will
happen. The previous code was designed to preserve existing semantics
in the face of shared locks, however, this was unnecessary.
Discussed with: mckusick
is requested. Handle this case specially before the while loop.
- Use the held vnode lock to check for VI_DOOMED. The vnode lock and
interlock must both be held to set VI_DOOMED so either one held, even
shared, is sufficient to check it.
No objection by: kib
are mixed. Some pure context switch microbenchmarks show up to 29%
improvement. Pipe based context switch microbenchmarks show up to 7%
improvement. Real world tests are far less impressive as they are
dominated more by actual work than switch overheads, but depending on
the machine in question, workload, kernel options, phase of moon, etc, a
few percent gain might be seen.
Summary of changes:
- don't reload MSR_[FG]SBASE registers when context switching between
non-threaded userland apps. These typically cost 120 clock cycles each
on an AMD cpu (less on Barcelona/Phenom). Intel cores are probably no
faster on this.
- The above change only helps unthreaded userland apps that tend to use
the same value for gsbase. Threaded apps will get no benefit from this.
- reorder things like accessing the pcb to be in memory order, to give
prefetching a better chance of working. Operations are now in increasing
memory address order, rather than reverse or random.
- Push some lesser used code out of the main code paths. Hopefully
allowing better code density in cache lines. This is probably futile.
- (part 2 of previous item) Reorder code so that branches have a more
realistic static branch prediction hint. Both Intel and AMD cpus
default to predicting branches to lower memory addresses as being
taken, and to higher memory addresses as not being taken. This is
overridden by the limited dynamic branch prediction subsystem. A trip
through userland might overflow this.
- Futule attempt at spreading the use of the results of previous operations
in new operations. Hopefully this will allow the cpus to execute in
parallel better.
- stop wasting 16 bytes at the top of kernel stack, below the PCB.
- Never load the userland fs/gsbase registers for kthreads, but preserve
curpcb->pcb_[fg]sbase as caches for the cpu. (Thanks Jeff!)
Microbenchmarking this code seems to be really sensitive to things like
scheduling luck, timing, cache behavior, tlb behavior, kernel options,
other random code changes, etc.
While it doesn't help heavy userland workloads much, it does help high
context switch loads a little, and should help those that involve
switching via kthreads a bit more.
A special thanks to Kris for the testing and reality checks, and Jeff for
tormenting me into doing this. :)
This is still work-in-progress.
PTE if that PTE has the PG_RW bit set. However, this assumption does
not hold on recent processors from Intel. For example, consider a PTE
that has the PG_RW bit set but the PG_M bit clear. Suppose this PTE
is cached in the TLB and later the PG_RW bit is cleared in the PTE,
but the corresponding TLB entry is not (yet) invalidated.
Historically, upon a write access using this (stale) TLB entry, the
TLB would observe that the PG_RW bit had been cleared and initiate a
page fault, aborting the setting of the PG_M bit in the PTE. Now,
however, P4- and Core2-family processors will set the PG_M bit before
observing that the PG_RW bit is clear and initiating a page fault. In
other words, the write does not occur but the PG_M bit is still set.
The real impact of this difference is not that great. Specifically,
we should no longer assert that any PTE with the PG_M bit set must
also have the PG_RW bit set, and we should ignore the state of the
PG_M bit unless the PG_RW bit is set. However, these changes enable
me to remove a work-around from pmap_promote_pde(), the superpage
promotion procedure.
(Note: The AMD processors that we have tested, including the latest,
the Phenom, still exhibit the historical behavior.)
Acknowledgments: After I observed the problem, Stephan (ups) was
instrumental in characterizing the exact behavior of Intel's recent
TLBs.
Tested by: Peter Holm
vnodes belonging to the mountpoint. Also, yield when in the
softdep_process_worklist() even when we are not going to sleep due to
buffer drain.
It is believed that the ULE fixed the problem [1], but the yielding
seems to be needed at least for the 4BSD case.
Discussed: on stable@, with bde
Reviewed by: tegge, jeff [1]
MFC after: 2 weeks
The overflow causes the wraparound with consequent corruption of the
(almost) whole address space mapping.
As Alan noted, pmap_copy() does not require the wrap-around checks
because it cannot be applied to the kernel's pmap. The checks there are
included for consistency.
Reported and tested by: kris (i386/pmap.c:pmap_remove() part)
Reviewed by: alc
MFC after: 1 week
multi-descriptor transmission attempt. Datasheet said nothing about
this requirements. This should fix a long-standing VLAN hardware
tagging issues with re(4).
Reported by: Giulio Ferro ( auryn AT zirakzigil DOT org )
Tested by: Giulio Ferro ( auryn AT zirakzigil DOT org )
to declaring a proper module. The module event handler is part of the
gpart core and will add the scheme to an internal list on module load
and will remove the scheme from the internal list on module unload.
This makes it possible to dynamically load and unload partitioning
schemes.
to it for tasting. This is useful when the class, through means outside
the scope of GEOM, can claim providers previously unclaimed.
The g_retaste() function posts an event which is handled by the
g_retaste_event().
Event suggested by: phk
exhaustion is encountered. There was a fix made previously for this
problem but the solution (breaking out of the receive loop) does not
seem to work. mbuf reuse strategy is already adopted by other drivers
such as if_bge. The problem was recreated and the patch is also
verified in the same test environment.
layouts different than the defaults:
o hint.npe.0.mac="A", "B", etc. specifies the window for MAC register accesses
o hint.npe.0.mii="A", "B", etc. specifies PHY registers
o hint.npe.1.phy=%d specifies the PHY to map to a port
This allows devices like NSLU to be setup w/o code changes and will
also be used for forthcoming support for more Avila boards.
Reviewed by: imp
MFC after 1 week
BO_LOCK/UNLOCK/MTX when manipulating the bufobj.
- Create a new lock in the bufobj to lock bufobj fields independently.
This leaves the vnode interlock as an 'identity' lock while the bufobj
is an io lock. The bufobj lock is ordered before the vnode interlock
and also before the mnt ilock.
- Exploit this new lock order to simplify softdep_check_suspend().
- A few sync related functions are marked with a new XXX to note that
we may not properly interlock against a non-zero bv_cnt when
attempting to sync all vnodes on a mountlist. I do not believe this
race is important. If I'm wrong this will make these locations easier
to find.
Reviewed by: kib (earlier diff)
Tested by: kris, pho (earlier diff)
code.
The bug:
There exists a race condition for timeout/untimeout(9) due to the
way that the softclock thread dequeues timeouts.
The softclock thread sets the c_func and c_arg of the callout to
NULL while holding the callout lock but not Giant. It then drops
the callout lock and acquires Giant.
It is at this point where untimeout(9) on another cpu/thread could
be called.
Since c_arg and c_func are cleared, untimeout(9) does not touch the
callout and returns as if the callout is canceled.
The softclock then tries to acquire Giant and likely blocks due to
the other cpu/thread holding it.
The other cpu/thread then likely deallocates the backing store that
c_arg points to and finishes working and hence drops Giant.
Softclock resumes and acquires giant and calls the function with
the now free'd c_arg and we have corruption/crash.
The fix:
We need to track curr_callout even for timeout(9) (LOCAL_ALLOC)
callouts. We need to free the callout after the softclock processes
it to deal with the race here.
Obtained from: Juniper Networks, iedowse
Reviewed by: jhb, iedowse
MFC After: 2 weeks.
around the check for the BV_BKGRDINPROG in the brelse() and bqrelse().
See the comment for the explanation why it is safe.
Tested by: pho
Submitted by: jeff
ffs_extread() when setting the IN_ACCESS flag by checking whether the
IN_ACCESS is already set. The possible race there is admissible.
Tested by: pho
Submitted by: jeff
