adrian [Mon, 28 Sep 2015 01:09:48 +0000 (01:09 +0000)]
if_otus fixes; add fast-frames support.
Fast-frames:
* include opt_wlan.h ; tsk to not doing it earlier;
* add a tx pending tracking counter for seeing how deep
the hardware TX queue is;
* add the frame aging code from if_ath;
* add fast-frames capability to the driver setup.
Bugs:
* free the buffers (and node references) before
detaching net80211 state. This prevents a use-after-free in
the node free path where we've destroyed net80211 underneath it.
adrian [Mon, 28 Sep 2015 00:59:07 +0000 (00:59 +0000)]
Migrate the fast-frames transmit support away from using the txa_private
field and into a separate fast-frames staging pointer in ieee80211_node.
The A-MPDU TX path allows txa_private to be used by drivers. So it will
clash with any attempt to use fast-frames. Now, fast-frames is not really
anything special - it's just a custom ethernet frame type that contains
two MSDUs into one MPDU. So all the NIC has to support doing is transmitting
up to a 4KiB frame with an arbitrary ethertype and bam! Fast-frames.
However, using txa_private means we can /either/ do fast-frames or A-MPDU TX,
so fast frames has been turned off in the Atheros HAL for 11n chipsets.
This is a bit silly - it actually means that 802.11 performance to/from
11abg Atheros chips is actually better than between an 11abg atheros device
and an 11n Atheros device.
So:
* create a new mbuf staging queue for fast frames. It only queues a single
frame in the staging queue (and there's a top-level ic staging queue
used for expiry/tracking) so it's just an mbuf pointer per TID.
* Still use the ampdu TX packet counter to determine whether to do
aggregation or not. It'll double count if we start doing both A-MPDU TX
and fast frames, but that's not all that important right now.
* Initialise the pps tracker so ticks isn't zero. This ensures that
fast-frames actually gets used - without it, the ticks math overflows
and the pps math always sets txa_pps=0. This is the same bug that
plagued A-MPDU TX starting logic.
This actually allows fast-frames transmit to occur between the AR9331
(in 11n HT/20 mode) and AR9170 (if_otus) in 11bg mode.
Now, this is a great big no-op on atheros 11n hardware, so don't worry.
It may mean you start seeing more reliable fast-frames transmission on
11abg hardware which may expose some more amusing bugs.
TODO:
* further testing and debugging of all of this before flipping on
fast-frames in if_ath (for 11n) and if_otus.
adrian [Mon, 28 Sep 2015 00:17:51 +0000 (00:17 +0000)]
Abstract out the ampdu TX pps initialisation code so it can be reused
in the superg fast-frames code.
This harks back to an earlier commit (r280349) where I found that
initialising the pps code with ticks=0 would cause hilariously bad
hz ticks wraparound failures, leading to never actually aggregating
traffic. This is still true for the superg path and so I have to
do the same thing there.
This is a big no-op; a subsequent commit will flip this on so it
works with the fast-frames transmit path.
Tested:
* AR9170, otus(4) - STA mode, 11bg operation
* AR9331, AP mode
fnmatch(): Remove exponential behaviour as in sh r229201.
The old code was exponential in the number of asterisks in the pattern.
However, once a match has been found upto the next asterisk, the previous
asterisks are no longer relevant.
Initially function was introduced in r53541 (KAME initial commit) to
"provide hints from upper layer protocols that indicate a connection
is making "forward progress"" (quote from RFC 2461 7.3.1 Reachability
Confirmation).
However, it was converted to do nothing (e.g. just return) in r122922
(tcp_hostcache implementation) back in 2003. Some defines were moved
to tcp_var.h in r169541. Then, it was broken (for non-corner cases)
by r186119 (L2<>L3 split) in 2008 (NULL ifp in nd6_lookup). So,
right now this code is broken and has no "real" base users.
jeff [Sun, 27 Sep 2015 05:16:06 +0000 (05:16 +0000)]
- Collapse vfs_vmio_truncate & vfs_vmio_release into a single function.
- Allow vfs_vmio_invalidate() to free the pages, leaving us with a
single loop and bufobj lock when B_NOCACHE/B_INVAL is used.
- Eliminate the special B_ASYNC handling on free that has not been
relevant for some time.
- Remove the extraneous page busy from vfs_vmio_truncate().
rtsock requests for deleting interface address lles started to return EPERM
instead of old "ignore-and-return 0" in r287789. This broke arp -da /
ndp -cn behavior (they exit on rtsock command failure). Fix this by
translating LLE_IFADDR to RTM_PINNED flag, passing it to userland and
making arp/ndp ignore these entries in batched delete.
adrian [Sun, 27 Sep 2015 04:03:11 +0000 (04:03 +0000)]
Enforce consistent limits of daemons run from rc.subr:
* Allow the user to configure the login class to use in rc.conf
by using {daemon}_login_class, which;
* Use the daemon class by default;
* .. and then use 'limits' to set the login class so it works both
via init at startup (which runs this in 'daemon' class) and via
whichever root environment (eg command line, other daemons, etc.)
The conversion of kmem_alloc_attr() from operating on a vm map to a vmem
arena in r254025 introduced a bug in the case when an allocation is only
partially successful. Specifically, the vm object lock was not being
acquired before freeing the allocated pages. To address this bug, replace
the existing code by a call to kmem_unback().
Change the type of a variable in kmem_alloc_attr() so that an allocation
of two or more gigabytes won't fail.
Replace the error handling code in kmem_back() by a call to kmem_unback().
Exploit r288122 to address a cosmetic issue. Since the pages allocated
by noobj_alloc() don't belong to a vm object, they can't be paged out.
Since they can't be paged out, they are never enqueued in a paging queue.
Nonetheless, passing PQ_INACTIVE to vm_page_unwire() creates the appearance
that these pages are being enqueued in the inactive queue. As of r288122,
we can avoid giving this false impression by passing PQ_NONE.
Hookup mkcsmapper_static and mkesdb_static for all but install.
These are only handled as 'build-tools' in Makefile.inc1. This causes
'make clean' from the top of the tree to not clean the directories. It also
effectively has kept them disconnected and risks them bitrotting. The
buildworld process never cleans them either.
Connect them so they will always be built, cleaned, etc, but never installed.
Hookup mkcsmapper_static and mkesdb_static for all but install.
These are only handled as 'build-tools' in Makefile.inc1. This causes
'make clean' from the top of the tree to not clean the directories. It also
effectively has kept them disconnected and risks them bitrotting. The
buildworld process never cleans them either.
Connect them so they will always be built, cleaned, etc, but never installed.
Make the __bitcount*() functions unconditionally available, by moving
them out of the #if _BSD_VISIBLE block. Other headers may depend on
__bitcount(). The dependencies can be a header not specified by
POSIX, and then namespace restrictions by _XOPEN_SOURCE are not
applicable, as it was reported. Or, we might grow an implementation
of some POSIX facility using __bitcount(), which also should work.
Reported by: Jason Schulz <schulz.j@gmail.com>
Discussed with: jhb
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
adrian [Sat, 26 Sep 2015 07:25:53 +0000 (07:25 +0000)]
* Add 11n HT40 support - i needed to send both HTINFO and HTCAP fields
in the join message so the firmware would pick it up.
* Strip out the direct hardware fiddling for 40MHz mode - the firmware
we're using doesn't require it (the rtl8712su firmware does; it
is less 'fullmac' than what we're using.)
* Fix the mbuf handling during errors - rsu_tx shouldn't free mbufs;
it's up to the caller to do so. This brings it in line with
what other drivers do or should be doing.
Tested:
* RTL8712, HT40 channel, STA mode (during this commit)
Exploit r288122 to address a cosmetic issue. Since PV chunk pages don't
belong to a vm object, they can't be paged out. Since they can't be paged
out, they are never enqueued in a paging queue. Nonetheless, passing
PQ_INACTIVE to vm_page_unwire() creates the appearance that these pages
are being enqueued in the inactive queue. As of r288122, we can avoid
this false impression by passing PQ_NONE.
adrian [Sat, 26 Sep 2015 07:08:35 +0000 (07:08 +0000)]
Add an initial driver for the AR9170 series draft-11n hardware from
Atheros.
Thanks to OpenBSD for providing a driver based on the original
Atheros open source driver circa 2008. This uses the early, pre-carl9170
atheros provided firmware.
It only supports 11bg at the moment. I've not tested it with 11a
(and so the TX rate control logic may be slightly wrong!) so if
you do have the dual-band version of this hardware please do let me know.
adrian [Sat, 26 Sep 2015 00:53:37 +0000 (00:53 +0000)]
Perform some rather amusing layering violations to add mbuf tags to the
net80211 receive path. This allows drivers (notably USB right now, but
anything/everything!) to optionally defer bulk RX of 802.11 frames until
/outside/ of the driver lock(s), rather than doing:
UNLOCK(sc);
ieee80211_input*()
LOCK(sc);
.. which is really stupid.
The existing API is maintaned - if ieee80211_input() / ieee80211_input_all()
is called then the RSSI/NF values are used. If the MIMO versions are called
with a given rx status pointer then it's used. Else, it'll use whatever
is in the RX mbuf tag.
Remove 'set -e' that are no longer needed as it is already default.
When bmake was initially imported at r241298 shell commands were no longer
ran with 'set -e' as they were before. This was fixed in r254980 so they
again always use 'set -e'.
Explicitly enable .MAKE.ALWAYS_PASS_JOB_QUEUE for bmake.
This is a NOP as r254419 enabled this by default in bmake. Add it here though
to ensure it is known that we are using this as a default and in case a
bmake import removes the default we have.
This tells bmake to always pass job tokens into sub-commands. Otherwise
it would only do so if the target being built depended on the special
.MAKE target (which causes _all_ commands to be executed with -n as well)
or if the command matches '${MAKE}/${.MAKE}/$(MAKE)/$(.MAKE)/make' (before
expansion, so ${LIB32WMAKE} would not qualify). Using '+' on a command
(which runs the command with -n) would not pass the job token even though it
is a documented way to achieve the .MAKE effect on a command.
Use per-cpu values for base and last in tc_cpu_ticks(). The values
are updated lockess, different CPUs write its own view of timecounter
state. The critical section is done for safety, callers of
tc_cpu_ticks() are supposed to already enter critical section, or to
own a spinlock.
The change fixes sporadical reports of too high values reported for
the (W)CPU on platforms that do not provide cpu ticker and use
tc_cpu_ticks(), in particular, arm*.
Diagnosed and reviewed by: jhb
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
MFV r288063: make dataset property de-registration operation O(1)
A change to a property on a dataset must be propagated to its descendants
in case that property is inherited. For datasets whose information is
not currently loaded into memory (e.g. a snapshot that isn't currently
mounted), there is nothing to do; the property change will take effect
the next time that dataset is loaded. To handle updates to datasets that
are in-core, ZFS registers a callback entry for each property of each
loaded dataset with the dsl directory that holds that dataset. There
is a dsl directory associated with each live dataset that references
both the live dataset and any snapshots of the live dataset. A property
change is effected by doing a traversal of the tree of dsl directories
for a pool, starting at the directory sourcing the change, and invoking
these callbacks.
The current implementation both registers and de-registers properties
individually for each loaded dataset. While registration for a property is
O(1) (insert into a list), de-registration is O(n) (search list and then
remove). The 'n' for de-registration, however, is not limited to the size
(number of snapshots + 1) of the dsl directory. The eviction portion
of the life cycle for the in core state of datasets is asynchronous,
which allows multiple copies of the dataset information to be in-core
at once. Only one of these copies is active at any time with the rest
going through tear down processing, but all copies contribute to the
cost of performing a dsl_prop_unregister().
One way to create multiple, in-flight copies of dataset information
is by performing "zfs list" operations from multiple threads
concurrently. In-core dataset information is loaded on demand and then
evicted when reference counts drops to zero. For datasets that are not
mounted, there is no persistent reference count to keep them resident.
So, a list operation will load them, compute the information required to
do the list operation, and then evict them. When performing this operation
from multiple threads it is possible that some of the in-core dataset
information will be reused, but also possible to lose the race and load
the dataset again, even while the same information is being torn down.
Compounding the performance issue further is a change made for illumos
issue 5056 which made dataset eviction single threaded. In environments
using automation to manage ZFS datasets, it is now possible to create
enough of a backlog of dataset evictions to consume excessive amounts
of kernel memory and to bog down the system.
The fix employed here is to make property de-registration O(1). With this
change in place, it is hoped that a single thread is more than sufficient
to handle eviction processing. If it isn't, the problem can be solved
by increasing the number of threads devoted to the eviction taskq.
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_dataset.c
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_dir.c:
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_prop.c:
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/dsl_dataset.h:
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/dsl_dir.h:
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/dsl_prop.h:
Associate dsl property callback records with both the
dsl directory and the dsl dataset that is registering the
callback. Both connections are protected by the dsl directory's
"dd_lock".
When linking callbacks into a dsl directory, group them by
the property type. This helps reduce the space penalty for the
double association (the property name pointer is stored once
per dsl_dir instead of in each record) and reduces the number of
strcmp() calls required to do callback processing when updating
a single property. Property types are stored in a linked list
since currently ZFS registers a maximum of 10 property types
for each dataset.
Note that the property buckets/records associated with a dsl
directory are created on demand, but only freed when the dsl
directory is freed. Given the static nature of property types
and their small number, there is no benefit to freeing the few
bytes of memory used to represent the property record earlier.
When a property record becomes empty, the dsl directory is either
going to become unreferenced a little later in this thread of
execution, or there is a high chance that another dataset is
going to be loaded that would recreate the bucket anyway.
Replace dsl_prop_unregister() with dsl_prop_unregister_all().
All callers of dsl_prop_unregister() are trying to remove
all property registrations for a given dsl dataset anyway. By
changing the API, we can avoid doing any lookups of callbacks
by property type and just traverse the list of all callbacks
for the dataset and free each one.
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_objset.c:
sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_vfsops.c:
Replace use of dsl_prop_unregister() with the new
dsl_prop_unregister_all() API.
illumos/illumos-gate@03bad06fbb261fd4a7151a70dfeff2f5041cce1f
Author: Justin Gibbs <gibbs@scsiguy.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Prakash Surya <prakash.surya@delphix.com>
Approved by: Dan McDonald <danmcd@omniti.com>
Illumos issue:
6171 dsl_prop_unregister() slows down dataset eviction
https://www.illumos.org/issues/6171
bsd.obj.mk handles the needs fine. When an objdir exists it will
just rm -Rf the objdir. When it does not exist though it will
call 'clean' and 'cleandepend', which properly recurse in bsd.progs.mk.
projects / FreeBSD / FreeBSD.git / log