The symptom: sometimes 11n (and non-11n) throughput is great.
Sometimes it isn't. Much teeth gnashing occured, and much kernel
bisecting happened, until someone figured out it was the order
of which things were rebooted, not the kernel versions.
(Which was great news to me, it meant that I hadn't broken if_ath.)
What we found was that sometimes the WME parameters for the best-effort
queue had a burst window ("txop") in which the station would be allowed
to TX as many packets as it could fit inside that particular burst
window. This improved throughput.
After initially thinking it was a bug - the WME parameters for the
best-effort queue -should- have a txop of 0, Bernard and I discovered
"aggressive mode" in net80211 - where the WME BE queue parameters
are changed if there's not a lot of high priority traffic going on.
The WME parameters announced in the association response and beacon
frames just "change" based on what the current traffic levels are.
So in fact yes, the STA was acutally supposed to be doing this higher
throughput stuff as it's just meant to be configuring things based on
the WME parameters - but it wasn't.
What was eventually happening was this:
* at startup, the wme qosinfo count field would be 0;
* it'd be parsed in ieee80211_parse_wmeparams();
* and it would be bumped (to say 10);
* .. and the WME queue parameters would be correctly parsed and set.
But then, when you restarted the assocation (eg hostap goes away and
comes back with the same qosinfo count field of 10, or if you
destroy the sta VIF and re-create it), the WME qosinfo count field -
which is associated not to the VIF, but to the main interface -
wouldn't be cleared, so the queue default parameters would be used
(which include no burst setting for the BE queue) and would remain
that way until the hostap qosinfo count field changed, or the STA
was actually rebooted.
This fix simply cleares the wme capability field (which has the count
field) to 0, forcing it to be reset by the next received beacon.
Thanks go to Milu for finding it and helping me track down what was
going on, and Bernard Schmidt for working through the net80211 and
WME specific magic.
Change BIO_GETATTR("GEOM::kerneldump") API to make set_dumper() called by
consumer (geom_dev) instead of provider (geom_disk). This allows any geom
insert it's code into the dump call chain, implementing more sophisticated
functionality then just disk partitioning.
queue has its own interrupt. If the exact number that we need is not a
power of 2 and we're using MSI, then switch to interrupt multiplexing.
While here, replace the magic numbers with something more readable.
MFC after: 3 days
At least one AR5416 user has reported measurable throughput drops
with this option. For now, disable it and make it a run-time
twiddle. It won't take affect until the next radio programming
trip though (eg channel scan, channel change.)
integer overflow when the input is very large (for example, 100 Pi would
become about 10 Ei which exceeded signed int64_t).
Solve this issue by splitting the division into two parts and avoid the
multiplication.
PR: bin/146205
Reviewed by: arundel
MFC after: 1 month
vfs_equalopts(). This allows vfs_sanitizeopts() to filter redundant
occurrences of these options. It was possible that for example both "ro"
and "rw" options became active concurrently.
PR: kern/133614
Discussed on: freebsd-hackers
MFC after: 1 month
Also, express this new maximum as a fraction of the kernel's address
space size rather than a constant so that increasing KVA_PAGES will
automatically increase this maximum. As a side-effect of this change,
kern.maxvnodes will automatically increase by a proportional amount.
While I'm here ensure that this change doesn't result in an unintended
increase in maxpipekva on i386. Calculate maxpipekva based upon the
size of the kernel address space and the amount of physical memory
instead of the size of the kmem map. The memory backing pipes is not
allocated from the kmem map. It is allocated from its own submap of
the kernel map. In short, it has no real connection to the kmem map.
(In fact, the commit messages for the maxpipekva auto-sizing talk
about using the kernel map size, cf. r117325 and r117391, even though
the implementation actually used the kmem map size.) Although the
calculation is now done differently, the resulting value for
maxpipekva should remain almost the same on i386. However, on amd64,
the value will be reduced by 2/3. This is intentional. The recent
change to VM_KMEM_SIZE_SCALE on amd64 for the benefit of ZFS also had
the unnecessary side-effect of increasing maxpipekva. This change is
effectively restoring maxpipekva on amd64 to its prior value.
Eliminate init_param3() since it is no longer used.
so there's no need to enable the RX of invalid frames just to do ANI.
The if_ath code and AR5212 ANI code setup the RX filter bits to enable
receiving OFDM/CCK errors if the device doesn't have the hardware
MIB counters. It isn't initialising it for the AR5416+ because all of
those chips have hardware MIB counters.
This fixes the odd (and performance affecting!) situation where if ani
is enabled (via sysctl dev.ath.X.intmit) then suddenly there's be a very
large volume of phy errors - which is good to track, but not what was
intended. Since each PHY error is a received (0 length) frame, it can
significantly tie up the RX side of things.
equal to secondary counters:
primary_localcnt = secondary_remotecnt
primary_remotecnt = secondary_localcnt
Previously it was done wrong and split-brain was observed after
primary had synchronized up-to-date data from secondary.
Approved by: pjd (mentor)
MFC after: 1 week
This (almost) gives us the address space back (at the bottom) that we lost
at the top.
Region 0 has traditionally been reserved for IA-32 emulation, which has not
been of great interest. By starting 64-bit processes at the 4G boundary we
at least preserve some of the advantages:
1. Any invalid pointer cast (from int to pointer and back) will still
always fail and not only when more than 4GB of memory is in use.
2. Memory sharing between 64-bit and 32-bit processes is still possibly
by using addresses < 4G.
