A default install on large memory machines with multiple 10gigE interfaces
were not being given enough mbufs to do full bandwidth TCP or NFS traffic.
To keep the value somewhat reasonable, we scale back the number of
maxuers by 1/6 past the 384 point. This gives us enough mbufs for most
of our pretty basic 10gigE line-speed tests to complete.
This enables CPU searches (which read tdq_load) to operate independently
of any contention on the spinlock. Some scheduler-intensive workloads
running on an 8C single-socket SNB Xeon show considerable improvement with
this change (2-3% perf improvement, 5-6% decrease in CPU util).
Sponsored by: Intel
Reviewed by: jeff
forked. Otherwise, pagedaemon might reclaim the page without saving
its content into the swap file, resulting in the valid content
replaced by zeroes.
Reported and tested by: pho
Reviewed and comment update by: alc
MFC after: 1 week
appear on which interface. This fixes detection of some USB audio adapters.
Also increase the channel limit for FULL speed devices to 4 channels.
Tested by: gavin
MFC after: 1 week
Also update the port reset time from 250ms to 50ms. Some USB devices
have a hard limit in hardware at 222ms for the port reset time and will
not enumerate unless this delay is closer to the usb.org defined value.
This patch can fix enumeration with some USB devices.
Tested by: Guido van Rooij
Submitted by: Nick Hibma
MFC after: 1 week
link at a lower speed so enabling it for fiber adapters is wrong.
Fix the issue by setting BGE_PHY_NO_WIRESPEED such that brgphy(4)
wouldn't enable the feature.
While I'm here move PHY specific feature/bug configuration to new
location(just before mii attach) for readability.
more appropriate named kernel options for the very distinct
send and receive path.
"options SOCKET_SEND_COW" enables VM page copy-on-write based
sending of data on an outbound socket.
NB: The COW based send mechanism is not safe and may result
in kernel crashes.
"options SOCKET_RECV_PFLIP" enables VM kernel/userspace page
flipping for special disposable pages attached as external
storage to mbufs.
Only the naming of the kernel options is changed and their
corresponding #ifdef sections are adjusted. No functionality
is added or removed.
Discussed with: alc (mechanism and limitations of send side COW)
before passing a packet to protocol input routines.
For several protocols this mean that now protocol needs to
do subtraction itself, and for another half this means that
we do not need to add header length back to the packet.
Make ip_stripoptions() to adjust ip_len, since now we enter
this function with a packet header whose ip_len does represent
length of entire packet, not payload only.
device drivers that used to provide this feature.
This is a subset of 241856 (which was reverted)
Reviewed by: des
Approved by: cperciva (implicit)
MFC after: 1 week
in network byte order. Any host byte order processing is
done in local variables and host byte order values are
never[1] written to a packet.
After this change a packet processed by the stack isn't
modified at all[2] except for TTL.
After this change a network stack hacker doesn't need to
scratch his head trying to figure out what is the byte order
at the given place in the stack.
[1] One exception still remains. The raw sockets convert host
byte order before pass a packet to an application. Probably
this would remain for ages for compatibility.
[2] The ip_input() still subtructs header len from ip->ip_len,
but this is planned to be fixed soon.
Reviewed by: luigi, Maxim Dounin <mdounin mdounin.ru>
Tested by: ray, Olivier Cochard-Labbe <olivier cochard.me>
In particular, do not lock Giant conditionally when calling into the
filesystem module, remove the VFS_LOCK_GIANT() and related
macros. Stop handling buffers belonging to non-mpsafe filesystems.
The VFS_VERSION is bumped to indicate the interface change which does
not result in the interface signatures changes.
Conducted and reviewed by: attilio
Tested by: pho
filters (ipfw and PF) use the same ruleset with the same lock for both
AF_INET and AF_INET6 there is no need in more fine-grade locking.
However, it is possible to request personal lock by specifying
PFIL_FLAG_PRIVATE_LOCK flag in pfil_head structure (see pfil.9 for
more details).
Export PFIL lock via rw_lock(9)/rm_lock(9)-like API permitting pfil consumers
to use this lock instead of own lock. This help reducing locks on main
traffic path.
pfil_assert() is currently not implemented due to absense of rm_assert().
Waiting for some kind of r234648 to be merged in HEAD.
This change is part of bigger patch reducing routing locking.
Sponsored by: Yandex LLC
Reviewed by: glebius, ae
OK'd by: silence on net@
MFC after: 3 weeks
are using (different) ND-based approach described in RFC 4861. This change
is similar to r241406 which conditionally skips the same check in IPv4.
This change is part of bigger patch eliminating rte locking.
Sponsored by: Yandex LLC.
OK'd by: hrs
MFC after: 2 weeks
Return EPERM if processes were found but they
were unable to be signaled.
Return the first error from p_cansignal if no signal was successful.
Reviewed by: jilles
Approved by: cperciva
MFC after: 1 week
Return EPERM if processes were found but they
were unable to be signaled.
Return the first error from p_cansignal if no signal was successful.
Reviewed by: jilles
Approved by: cperciva
MFC after: 1 week
.. so that consistent compilation algorithms are used for both
architectures as in practice the binaries are expected to be
interchangeable (for time being).
Previously i386 used default setting which were equivalent to
-march=i486 -mtune=generic.
The only difference is using smaller but slower "leave" instructions.
Discussed with: jhb, dim
MFC after: 29 days
output and replace it with a new visible sysctl kern.ipc.acceptqueue
of the same functionality. It specifies the maximum length of the
accept queue on a listen socket.
The old kern.ipc.somaxconn remains available for reading and writing
for compatibility reasons so that existing programs, scripts and
configurations continue to work. There no plans to ever remove the
orginal and now hidden kern.ipc.somaxconn.
The reason for this is that the SPARC v9 architecture allows nested
interrupts of higher priority/level than that of the current interrupt
to occur (and we can't just entirely bypass this model, also, at least
for tick interrupts, this also wouldn't be wise). However, when a
preemption interrupt interrupts another interrupt of lower priority,
f.e. PIL_ITHREAD, and that one in turn is nested by a third interrupt,
f.e. PIL_TICK, with SCHED_ULE the execution of interrupts higher than
PIL_PREEMPT may be migrated to another CPU. In particular, tl1_ret(),
which is responsible for restoring the state of the CPU prior to entry
to the interrupt based on the (also migrated) trap frame, then is run
on a CPU which actually didn't receive the interrupt in question,
causing an inappropriate processor interrupt level to be "restored".
In turn, this causes interrupts of the first level, i.e. PIL_ITHREAD
in the above scenario, to be blocked on the target of the migration
until the correct PIL happens to be restored again on that CPU again.
Making PIL_PREEMPT the lowest real priority, this effectively prevents
this scenario from happening, as preemption interrupts no longer can
interrupt any other interrupt besides stray ones (which is no issue).
Thanks to attilio@ and especially mav@ for helping me to understand
this problem at the 201208DevSummit.
- Give PIL_STOP (which is also used for IPI_STOP_HARD, given that there's
no real equivalent to NMIs on SPARC v9) the highest possible priority
just below the hardwired PIL_TICK, so it has a chance to interrupt
more things.
MFC after: 1 week
... otherwise the current thread might be holding ARC locks and thus run
into a deadlock. This happens, for example, when a thread does memory
allocation in the ARC code and runs into KVA shortage.
Also, it really makes the most sense to wait in pageproc, so that the
results of ARC reclamation are seen before the page cache is acted upon.
In other cases where vm_lowmem is invoked, e.g. on KVA space shortage,
the callers perform multiple attempts (up to 8) and wait for rather
long intervals between them (up to 4 seconds), so ARC reclaim results
should become visible even without explicit waiting on the ARC thread.
Note that this is not a critical issue for typical ZFS usages where KVA
space should already be large enough. On amd64 systems setting KVA size
to twice the physical memory size is known to mitigate KVA fragmentation
issues in practice.
Side note: perhaps vm_lowmem 'how' parameter should be used to
differentiate between causes of the event.
Reported by: Nikolay Denev <ndenev@gmail.com>
MFC after: 19 days
parent adapter's _DOD list, only check the low 16 bits of both _ADR and
_DOD. The language in the ACPI spec seems to indicate that the _ADR values
should exactly match the entries in _DOD. However, I assume that the
masking added to _DOD values was added to work around some known busted
machines (the commit history doesn't indicate either way), and the ACPI
spec does require that the low 16 bits are unique for all video outputs,
so only check the low 16 bits should be fine.
This fixes recognition of video outputs that use the new standardized
device ID scheme in ACPI 3.0 that set bit 31 such as certain Dell laptops.
Tested by: Juergen Lock nox jelal kn-bremen de
MFC after: 3 days
(Model 0x2D /* Per Intel document 253669-044US 08/2012. */)
Add manpage to document all the goodness that is available in this
processor model.
No support for uncore events at this time.
Submitted by: hiren panchasara <hiren.panchasara@gmail.com>
Reviewed by: jimharris@ fabient@
Obtained from: Yahoo! Inc.
MFC after: 2 weeks
when running tick_process(), similarly to what the x86 equivalents of
this function do, however employing the less racy sequence also used in
intr_event_handle().
MFC after: 3 days
that revises the netmap memory allocator so that the
various parameters (number and size of buffers, rings, descriptors)
can be modified at runtime through sysctl variables.
The changes become effective when no netmap clients are active.
The API is mostly unchanged, although the NIOCUNREGIF ioctl now
does not bring the interface back to normal mode: and you
need to close the file descriptor for that.
This change was necessary to track who is using the mapped region,
and since it is a simplification of the API there was no
incentive in trying to preserve NIOCUNREGIF.
We will remove the ioctl from the kernel next time we need
a real API change (and version bump).
Among other things, buffer allocation when opening devices is
now much faster: it used to take O(N^2) time, now it is linear.
Submitted by: Giuseppe Lettieri
GIANT from VFS. In addition, disconnect also netsmb, which is a base
requirement for SMBFS.
In the while SMBFS regular users can use FUSE interface and smbnetfs
port to work with their SMBFS partitions.
Also, there are ongoing efforts by vendor to support in-kernel smbfs,
so there are good chances that it will get relinked once properly locked.
This is not targeted for MFC.
doesn't automatically clear when VDD rises above Vlow again and needs to be
cleared manually. However, apparently this needs all of the time registers
to be set, i.e. pcf8563_settime(), and not just PCF8563_R_SECOND in order
for PCF8563_R_SECOND_VL to stick. Thus, we just issue a warning during
pcf8563_attach() rather than failing with ENXIO in case it is set.
MFC after: 3 days
This eliminates the need to manage queue depth at the nvd(4) level for
Chatham prototype board workarounds, and also adds the ability to
accept a number of requests on a single qpair that is much larger
than the number of trackers allocated.
Sponsored by: Intel
nvme_ctrlr_submit_io_request().
While here, also fix case where a uio may have more than 1 iovec.
NVMe's definition of SGEs (called PRPs) only allows for the first SGE to
start on a non-page boundary. The simplest way to handle this is to
construct a temporary uio for each iovec, and submit an NVMe request
for each.
Sponsored by: Intel
from an NVMe consumer.
This allows us to mostly build NVMe command buffers without holding the
qpair lock, and also allows for future queueing of nvme_request objects
in cases where the submission queue is full and no nvme_tracker objects
are available.
Sponsored by: Intel
This simplifies the driver significantly where it is constructing
commands to be submitted to hardware. By reducing the number of
PRPs (NVMe parlance for SGE) from 128 to 32, it ensures we do not
allocate too much memory for more common smaller I/O sizes, while
still supporting up to 128KB I/O sizes.
This also paves the way for pre-allocation of nvme_tracker objects
for each queue which will simplify the I/O path even further.
Sponsored by: Intel
GIANT from VFS. This code is particulary broken and fragile and other
in-kernel implementations around, found in other operating systems,
don't really seem clean and solid enough to be imported at all.
If someone wants to reconsider in-kernel NTFS implementation for
inclusion again, a fair effort for completely fixing and cleaning it
up is expected.
In the while NTFS regular users can use FUSE interface and ntfs-3g
port to work with their NTFS partitions.
This is not targeted for MFC.
GIANT from VFS. In addition, disconnect also netncp, which is a base
requirement for NWFS.
In the possibility of a future maintenance of the code and later
readd to the FreeBSD base, maybe we should think about a better location
for netncp. I'm not entirely sure the / top location is actually right,
however I will let network people to comment on that more specifically.
This is not targeted for MFC.
getnewvnode_reserve helps to avoid "recursing" back into zfs code
via getnewvnode when that latter needs to reclaim some vnodes.
zfs code may hold a number of locks around getnewvnode and doesn't
expect any recursion to happen on those locks, because that never
happens in solaris.
I believe that this change also eleiminates a need for the delayed
znode destruction via the taskqueue.
Many thanks to kib for devising getnewvnode_reserve.
Reported by: flo
Tested by: bapt, kwm, swills
MFC after: 2 weeks
X-MFC after: r241556
around the problem where high speed interfaces (such as ixgbe(4))
are not able to report real ifi_baudrate. bascially, take a spare
byte from struct if_data and use it to store ifi_baudrate power
factor. in other words,
real ifi_baudrate = ifi_baudrate * 10 ^ ifi_baudrate power factor
this should be backwards compatible with old binaries. use ixgbe(4)
as an example on how drivers would set ifi_baudrate power factor
Discussed with: kib, scottl, glebius
MFC after: 1 week
now use function calls:
if_clone_simple()
if_clone_advanced()
to initialize a cloner, instead of macros that initialize if_clone
structure.
Discussed with: brooks, bz, 1 year ago
sdchi encapsulates a generic SD Host Controller logic that relies on
actual hardware driver for register access.
sdhci_pci implements driver for PCI SDHC controllers using new SDHCI
interface
No kernel config modifications are required, but if you load sdhc
as a module you must switch to sdhci_pci instead.
- Use device_printf() and device_get_unit() instead of storing the unit
number in the softc.
- Remove use of explicit bus space handles and tags.
- Remove the global dpt_softcs list and use devclass_get_device() instead.
- Use pci_enable_busmaster() rather than frobbing the PCI command register
directly.
Tested by: no one
- Use device_printf() and device_get_unit() instead of storing the unit
number in the softc.
- Remove use of explicit bus space handles and tags.
- Return an errno value from bt_eisa_attach() if an error occurs rather
than -1.
- Use BUS_PROBE_DEFAULT rather than 0.
Tested by: no one
- Move 'free_scbs' into the softc rather than having it be a global list
and convert it to an SLIST instead of a hand-rolled linked-list.
- Use device_printf() and device_get_unit() instead of storing the unit
number in the softc.
- Remove use of explicit bus space handles and tags.
- Don't call device_set_desc() in the pccard attach routine, instead
set a default description during the pccard probe if the matching
product doesn't have a name.
Tested by: no one
- Use device_printf() and device_get_unit() instead of storing the unit
number in the softc.
- Remove use of explicit bus space handles and tags.
- Compare pointers against NULL.
- Let new-bus allocate a softc rather than doing it by hand.
Tested by: no one
- Use device_printf() and device_get_nameunit() instead of adw_name().
- Remove use of explicit bus space handles and tags.
- Use pci_enable_busmaster() rather than frobbing the PCI command register
directly.
- Use the softc provided by new-bus rather than allocating a new one.
Tested by: no one
ip6_output(), the IPv6 stack is working in net byte order.
The reason this code worked before is that ip6_output()
doesn't look at ip6_plen at all and recalculates it based
on mbuf length.
* Record TX mbufs when we get them so we can release them.
* Set TX/RX mbuf slots to NULL when we are no longer responsible for them
* Move dma sync on RX into RX intr routine
stashed away in ath_node.
As much as I tried to stuff that behind the ATH_NODE lock, unfortunately
the locking is just too plain hairy (for me! And I wrote it!) to do
cleanly. Hence using atomics here instead of a lock. The ATH_NODE lock
just isn't currently used anywhere besides the rate control updates.
If in the future everything gets migrated back to using a single ATH_NODE
lock or a single global ATH_TX lock (ie, a single TX lock for all TX and
TX completion) then fine, I'll remove the atomics.
processes running as root to suspend/resume execution
of the kernel nfsd threads. An earlier version of this
patch was tested by Vincent Hoffman (vince at unsane.co.uk)
and John Hickey (jh at deterlab.net).
Reviewed by: kib
MFC after: 2 weeks
it run out of multiple concurrent contexts.
Right now the ath(4) TX processing is a bit hairy. Specifically:
* It was running out of ath_start(), which could occur from multiple
concurrent sending processes (as if_start() can be started from multiple
sending threads nowdays.. sigh)
* during RX if fast frames are enabled (so not really at the moment, not
until I fix this particular feature again..)
* during ath_reset() - so anything which calls that
* during ath_tx_proc*() in the ath taskqueue - ie, TX is attempted again
after TX completion, as there's now hopefully some ath_bufs available.
* Then, the ic_raw_xmit() method can queue raw frames for transmission
at any time, from any net80211 TX context. Ew.
This has caused packet ordering issues in the past - specifically,
there's absolutely no guarantee that preemption won't occuring _during_
ath_start() by the TX completion processing, which will call ath_start()
again. It's a mess - 802.11 really, really wants things to be in
sequence or things go all kinds of loopy.
So:
* create a new task struct for TX'ing;
* make the if_start method simply queue the task on the ath taskqueue;
* make ath_start() just be called by the new TX task;
* make ath_tx_kick() just schedule the ath TX task, rather than directly
calling ath_start().
Now yes, this means that I've taken a step backwards in terms of
concurrency - TX -and- RX now occur in the same single-task taskqueue.
But there's nothing stopping me from separating out the TX / TX completion
code into a separate taskqueue which runs in parallel with the RX path,
if that ends up being appropriate for some platforms.
This fixes the CCMP/seqno concurrency issues that creep up when you
transmit large amounts of uni-directional UDP traffic (>200MBit) on a
FreeBSD STA -> AP, as now there's only one TX context no matter what's
going on (TX completion->retry/software queue,
userland->net80211->ath_start(), TX completion -> ath_start());
but it won't fix any concurrency issues between raw transmitted frames
and non-raw transmitted frames (eg EAPOL frames on TID 16 and any other
TID 16 multicast traffic that gets put on the CABQ.) That is going to
require a bunch more re-architecture before it's feasible to fix.
In any case, this is a big step towards making the majority of the TX
path locking irrelevant, as now almost all TX activity occurs in the
taskqueue.
Phew.
Right now processing a full 512 frame queue takes quite a while (measured
on the order of milliseconds.) Because of this, the TX processing ends up
sometimes preempting the taskqueue:
* userland sends a frame
* it goes in through net80211 and out to ath_start()
* ath_start() will end up either direct dispatching or software queuing a
frame.
If TX had to wait for RX to finish, it would add quite a few ms of
additional latency to the packet transmission. This in the past has
caused issues with TCP throughput.
Now, as part of my attempt to bring sanity to the TX/RX paths, the first
step is to make the RX processing happen in smaller 'parts'. That way
when TX is pushed into the ath taskqueue, there won't be so much latency
in the way of things.
The bigger scale change (which will come much later) is to actually
process the frames in the ath_intr taskqueue but process _frames_ in
the ath driver taskqueue. That would reduce the latency between
processing and requeuing new descriptors. But that'll come later.
The actual work:
* Add ATH_RX_MAX at 128 (static for now);
* break out of the processing loop if npkts reaches ATH_RX_MAX;
* if we processed ATH_RX_MAX or more frames during the processing loop,
immediately reschedule another RX taskqueue run. This will handle
the further frames in the taskqueue.
This should have very minimal impact on the general throughput case,
unless the scheduler is being very very strange or the ath taskqueue
ends up spending a lot of time on non-RX operations (such as TX
completion.)
counter, without actually allocating the vnodes. The supposed use of
the getnewvnode_reserve(9) is to reclaim enough free vnodes while the
code still does not hold any resources that might be needed during the
reclamation, and to consume the slack later for getnewvnode() calls
made from the innards. After the critical block is finished, the
caller shall free any reserve left, by getnewvnode_drop_reserve(9).
Reviewed by: avg
Tested by: pho
MFC after: 1 week
and Sierra Wireless MC8790V. Also implement the .ucom_poll method.
Note: This makes it possible to use lqr/echo in ppp.conf. And it
resolves ppp hanging during the PPp> phase.
Reviewed by: hps
MFC after: 1 week
AMD BKDG for CPU families 10h and later requires that the memory
mapped config is always read into or written from al/ax/eax register.
Discussed with: kib, alc
Reviewed by: kib (earlier version)
MFC after: 25 days
System time is set later on boot process then initial bus scan by CAM.
Until that moment microtime() is equal to microuptime(), and if system
boots quickly, the value can be close to zero. That causes settle time
waiting even for buses that don't use reset during probe.
On my test system this reduces boot time by 1 second if USB enabled, or
by 4 seconds if USB disabled. CAM waited for ctl2cam0 bus "settle".
"boot verbose", "single user mode", "ACPI" and more are now stateful boolean
menuitems rather than direct action-items.
A short-coming in this new menu system is that when a user sets a non-default
value in loader.conf(5), this non-default state is not reflected in the menu
-- leading to confusion as to whether the option was taking effect or not.
This patch adds dynamic menuitem constructors _and_ the necessary Forth
callbacks to initialize these stateful menuitems -- causing the aforementioned
menuitems to adhere to loader.conf(5) settings.
PR: bin/172529
Approved by: adrian (co-mentor)
MFC after: 21 days
This has been developed during 2 summer of code mandates and being revived
by gnn recently.
The functionality in this commit mirrors entirely content of fusefs-kmod
port, which doesn't need to be installed anymore for -CURRENT setups.
In order to get some sparse technical notes, please refer to:
http://lists.freebsd.org/pipermail/freebsd-fs/2012-March/013876.html
or to the project branch:
svn://svn.freebsd.org/base/projects/fuse/
which also contains granular history of changes happened during port
refinements. This commit does not came from the branch reintegration
itself because it seems svn is not behaving properly for this functionaly
at the moment.
Partly Sponsored by: Google, Summer of Code program 2005, 2011
Originally submitted by: ilya, Csaba Henk <csaba-ml AT creo DOT hu >
In collabouration with: pho
Tested by: flo, gnn, Gustau Perez,
Kevin Oberman <rkoberman AT gmail DOT com>
MFC after: 2 months
vm_page_sleep(). vm_page_sleep() is no longer called with this lock
held.
Eliminate assertions that the page queues lock is NOT held. These
assertions won't translate well to having distinct locks on the active
and inactive page queues, and they really aren't that useful.
MFC after: 3 weeks
- Extend the lock to cover xpt_path_release() for the new path.
- While xpt_action() is called while holding right SIM lock for the new
bus, the old path release may require different SIM lock. So we have
to temporary drop the new lock and get the old one.
without holding SIM lock. It really doesn't need that lock, but adding it
removes that specific exception, allowing to assert locking there later.
Submitted by: ken@ (earlier version)
- Disable the support for the second channel on twin-channel EISA cards as
the current incarnation can't possibly work correctly (it hasn't worked
since switching to new-bus where new-bus allocates the softc). If anyone
bothers to test this again it can be fixed properly and brought back.
- Use device_printf() and device_get_nameunit() instead of adv_name().
- Remove use of explicit bus space handles and tags.
- Use PCI bus accessors and helper routines rather than accessing
config registers directly.
- Handle failures from adv_attach().
Tested by: no one (hope it works)
