adrian [Mon, 21 Sep 2015 02:30:22 +0000 (02:30 +0000)]
Convert if_rsu to use a deferred transmit task rather than using rsu_start()
to do it directly.
Ensure that we re-queue starting transmit upon TX completion.
This solves two issues:
* It stops tx stalls - before this, if the transmit path filled the
mbuf queue then it'd never start another transmit.
* It enforces ordering - this is very required for 802.11n which
requires frames to be transmitted in the order they're queued.
Since everything remotely involved in USB has an unlock/thing/relock
pattern with that mutex, the only way to guarantee TX ordering is
to 100% defer it into a separate thread.
This now survives an iperf test and gets a reliable 30mbit/sec.
adrian [Mon, 21 Sep 2015 02:12:01 +0000 (02:12 +0000)]
Drain the mbuf queue upon rsu_stop().
Correctly (I hope!) remove net80211 references before doing so.
Just doing a dumb mbufq drain isn't enough.
If enough traffic occurs and the mbuf queue fills up then transmit
stalls (which I'm not fixing in this commit!) but then the mbuf queue
stays full until the driver is removed. There's also the net80211
node refcounting leak.
This just ensures that during rsu_stop and detach the mbuf queue
is purged (and references!) so the queue-full situation can be
recovered from.
Ensure that when a blockable open of fifo returns success, a valid
file descriptor opened for complimentary access exists as well.
The implementation of the guarantee is done by counting the
generations of readers and writers opens. We return success and not
EINTR or ERESTART error, when the sleep for complimentary opening is
interrupted, but the generation was changed during the sleep.
Longer explanation: assume there are two threads, A doing open("fifo",
O_RDONLY) and B doing open("fifo", O_WRONLY), and no other threads
either trying to open the fifo, nor there are any file descriptors
referencing the fifo. Before the change, it was possible e.g. for for
thread A to return a valid file descriptor, while thread B returned
EINTR if a signal to B was delivered simultaneously with the wakeup
from A. After the change, in this situation both A::open() and
B::open() succeed and the signal is made "as if" it was noticed
slightly later. Note that the signal actual delivery is not changed,
it is done by ast on syscall return path, so signal handler is still
executed before first instruction after syscall.
See PR for the code demonstrating the issue.
PR: 203162
Reported by: Victor Stinner victor.stinner@gmail.com
Reviewed by: jilles
Tested by: bapt, pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Correct a non-fatal error in vm_pageout_worker(). vm_pageout_worker()
should not assume that vm_pages_needed will remain set while it sleeps.
Other threads can clear vm_pages_needed by performing a sufficient
number of vm_page_free() calls, e.g., process termination. The effect
of this error was that vm_pageout_worker() would free and/or launder
pages when, in fact, there was no shortage of free pages.
Rewrite a nearby comment to describe all of the possible cases and not
just the most common case. The problem being that the comment made
the most common case seem like the only case.
Add support for weak symbols to the kernel linkers. It means that
linkers no longer raise an error when undefined weak symbols are
found, but relocate as if the symbol value was 0. Note that we do not
repeat the mistake of userspace dynamic linker of making the symbol
lookup prefer non-weak symbol definition over the weak one, if both
are available. In fact, kernel linker uses the first definition
found, and ignores duplicates.
Signature of the elf_lookup() and elf_obj_lookup() functions changed
to split result/error code and the symbol address returned.
Otherwise, it is impossible to return zero address as the symbol
value, to MD relocation code. This explains the mechanical changes in
elf_machdep.c sources.
The powerpc64 R_PPC_JMP_SLOT handler did not checked error from the
lookup() call, the patch leaves the code as is (untested).
Reported by: glebius
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Add PF_LOCAL support in getaddrinfo(3) and getnameinfo(3):
- In a PF_LOCAL address, "hostname" must begins with '/' and "servname"
is always NULL. All of ai_flags are ignored.
- PF_UNSPEC matches PF_LOCAL. EAI_SERVICE is not returned to make
AF-independent programming easier; "servname" is always ignored
in PF_LOCAL. In practice, PF_INET* and PF_LOCAL are
mutually-exclusive because a hostname which begins with '/' is invalid
in PF_INET*. No domain name resolution is performed for a PF_LOCAL address.
- Use NI_MAXHOST-long buffer for getnameinfo().
Although INET6_ADDRSTRLEN was designed to hold the longest
IPv6 address in IPv4-mapped address format a long time ago,
getnameinfo() can return scope identifier in addition to it.
- Fix a crash on a rpc entry when an IPv6 address is explicitly specified
in -a flag.
- Fix a bug that sockaddr_in was used where sockaddr_in6 should have
been used. This was not actually harmful because offsetof(struct
sockaddr_in, sin_port) is equal to offsetof(struct sockaddr_in6,
sin6_port).
'sin' is never used after assignment. Looking at the context, it seems
that it belongs the commented out section of code so make it part of that
section.
Always execute REPORT LUNS and REQUEST SENSE commands locally.
REPORT LUNS command is more related to target rather then specific LUN.
This node may be primary for LUNs for some reason unknown to another,
and command forwarded to another node won't be able to report them.
REQUEST SENSE is related to LUN, but in our implementation it reports
only UAs and CAs, that are stored locally rather then on primary node.
- Return an error if no matching device is found when the locate command is run
- Enhance the locate command to be able to address drive bays with no disk, or where the SES controller has not made the mapping to the device name
- Added the fault command, similar to locate, but a different SES property. On some of my controllers locate blinks the activity light, others the fault light. The fault command keeps the fault light on constant.
- Improve the usage() output and use it everywhere
- Added the map command, displays all elements connected to each (or the specified) ses(4) controller
- Added the status command, returns the overall status of the ses(4) controller
Cleanup nd6_cache_lladdr(). No functional changes.
* Since new extries are now allocated explicitly, fill in
all the necessary fields for lle _before_ attaching it to the table.
* Remove ND6_LLINFO_INCOMPLETE check which was unused even in
first KAME merge (r53541).
* After that, the only new state that function can set, was
ND6_LLINFO_STALE. Given everything above, simplify logic besides
do_update and is_newentry.
* Fix nd_resolve() comment.
The exists(${DESTDIR}...) check runs with DESTDIR being blank. When the
target runs it does have DESTDIR=${STAGE_OBJTOP} via bsd.sys.mk. This
results in the first execution warning that the symlink is missing. The
second run does run fine. However, this chflags is not needed at all
for META_MODE/STAGING since we never had this path being a schg file
while using META_MODE.
If the command to be ran changes then a rebuild is caused. Checking
exists(${DESTDIR}...) from make results in this on the 2nd and
possibly subsequent builds due to staging during build. Avoid this
by always running the existence check in the make sh command.
Previously, with serseq enabled, next command was unblocked only after
previous completed. With this change, for read operations, next command
is unblocked as soon as last media read completed. This is important
for frontends that actually wait for data move completion (like camtgt),
or when data are moved through the HA link, or especially when both.
Kernel part of reroot support - a way to change rootfs without reboot.
Note that the mountlist manipulations are somewhat fragile, and not very
pretty. The reason for this is to avoid changing vfs_mountroot(), which
is (obviously) rather mission-critical, but not very well documented,
and thus hard to test properly. It might be possible to rework it to use
its own simple root mount mechanism instead of vfs_mountroot().
Reviewed by: kib@
MFC after: 1 month
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D2698
Do not execute exception handlers with disabled interrupts.
We should not call vm_fault(), or send a signal, with interrupts
disabled. MI kernel code is not prepared for such environment, not to
mention that this increases system latency, since code appears to be
executing as being under spinlock.
The FAR register for data aborts is read before the interrupts are
enabled, to avoid its corruption due to nested exception or context
switch.
Add asserts, similar to the checks done by other architectures, about
not taking page faults in non-sleepable contexts, rather than die with
late and somewhat confusing witness diagnostic.
Reviewed by: andrew
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D3669
adrian [Fri, 18 Sep 2015 07:55:33 +0000 (07:55 +0000)]
Add in a temporary (hah!) workaround for net80211 scanning versus NIC
requirements.
Don't start the opmode and join path until a pending survey is finished.
This seems to reliably fix things.
Ideally I'd just finish off the net80211 pluggable scan stuff and implement
the methods here so if_rsu can just drive the scan machinery.
However, that's a .. later thing.
Whilst here, remove the getbuf debugging; it's okay to run out of transmit
buffers under load; it however isn't okay to not be able to send commands.
I'll fix that later.
adrian [Fri, 18 Sep 2015 05:01:05 +0000 (05:01 +0000)]
Add an external facing function to manually set the RX A-MPDU parameters
for re-ordering.
Devices like if_rsu don't pass through action/management frames but do send
firmware commands to inform us of things. One of those notifications is
the RX A-MPDU negotiated parameters.
adrian [Fri, 18 Sep 2015 04:12:11 +0000 (04:12 +0000)]
Add initial 11n support to if_rsu.
* Add a tunable to enable 11n if it's available, so to not anger people
who upgrade.
kenv hw.usb.rsu.enable_11n=1 before inserting the device.
* Add initial 11n htconfig bits;
* Enable 40MHz mode if it's available;
* Add 11n channels;
* Set 11n bits in the firmware.
It works for RX; I haven't tested TX aggregation just yet.
However the firmware doesn't do RX re-ordering, so I have to tie it into
the net80211 A-MPDU RX reorder path before I flip this on by default.
I've verified that I'm indeed actually seeing MCS 0->7 rates being received.
I haven't dug into whether it's actually transmitting 11n rates; I'll dig into
that later.