Alexander Motin [Wed, 13 Apr 2011 06:36:22 +0000 (06:36 +0000)]
As soon as siis_reset() doesn't waits for device readiness, but only for
controller port readiness (that should set just after PHY ready signal),
reduce wait time from 10s to 1s before trying more aggressive reset method.
This should improve system responsibility in some failure conditions.
Adrian Chadd [Wed, 13 Apr 2011 04:40:59 +0000 (04:40 +0000)]
Port over a TX gain fix from ath9k specific to the AR9285 (Kite) and AR9271.
Note: this HAL currently only supports the AR9285.
From Linux ath9k:
The problem is that when the attenuation is increased,
the rate will start to drop from MCS7 -> MCS6, and finally
will see MCS1 -> CCK_11Mbps. When the rate is changed b/w
CCK and OFDM, it will use register desired_scale to calculate
how much tx gain need to change.
The output power with the same tx gain for CCK and OFDM modulated
signals are different. This difference is constant for AR9280
but not AR9285/AR9271. It has different PA architecture
a constant. So it should be calibrated against this PA
characteristic.
The driver has to read the calibrated values from EEPROM and set
the tx power registers accordingly.
- Fix the code that matches userids in match_jobspec(). It needs to check
the username-for-accounting field (P), not the username-for-headerpage (L).
These are usually the same value, except that control files do not have
the username-for-headerpage field if the user has requested no header page.
- Also rename the cji_username field to cji_headruser, to make it clear that
the value should only be used for the header page. (aka banner page)
Rick Macklem [Wed, 13 Apr 2011 00:03:49 +0000 (00:03 +0000)]
Fix a couple of mbuf leaks introduced by r217242. I do
not believe that these leaks had a practical impact,
since the situations in which they would have occurred
would have been extremely rare.
Jung-uk Kim [Tue, 12 Apr 2011 23:04:01 +0000 (23:04 +0000)]
Reinstate cpu_est_clockrate() support for P-state invariant TSC if APERF and
MPERF MSRs are available. It was disabled in r216443. Remove the earlier
hack to subtract 0.5% from the calibrated frequency as DELAY(9) is little
bit more reliable now.
Jung-uk Kim [Tue, 12 Apr 2011 22:41:52 +0000 (22:41 +0000)]
Do not use TSC for DELAY(9) if it not P-state invariant to avoid possible
foot-shooting. DELAY() becomes unreliable when TSC frequency varies wildly,
especially cpufreq(4) and powerd(8) are used at the same time.
Jung-uk Kim [Tue, 12 Apr 2011 21:08:34 +0000 (21:08 +0000)]
Add a new tunable 'machdep.disable_tsc_calibration' to allow skipping TSC
frequency calibration. For Intel processors, if brand string from CPUID
contains its nominal frequency, this frequency is used instead.
Alexander Motin [Tue, 12 Apr 2011 20:50:57 +0000 (20:50 +0000)]
Refactor hard-reset implementation in ahci(4).
Instead of spinning in a tight loop for up to 15 seconds, polling for device
readiness while it spins up, return reset completion just after PHY reports
"connect well" or 100ms connection timeout. If device was found, use callout
for checking device readiness with 100ms period up to full 31 second timeout.
This fixes system freeze for 5-10 seconds on drives hot plug-in.
Alexander Motin [Tue, 12 Apr 2011 09:55:24 +0000 (09:55 +0000)]
Implement automatic SCSI sense fetching for ata(4) in ATA_CAM mode.
While it could be successfully done by CAM error recovery code, I was
told by several people that it is also a SIM obligation.
Adrian Chadd [Tue, 12 Apr 2011 08:10:25 +0000 (08:10 +0000)]
Introduce geom_map, a GEOM provider designed for use by
embedded flash stores.
Some devices - notably those with uboot - don't have an
explicit partition table (eg like Redboot's FIS.)
geom_map thus provides an easy way to export the hard-coded
flash layout as geom providers for use by filesystems and
other tools.
It also includes a "search" function which allows for
dynamic creation of partition layouts where the device only
has a single hard-coded partition. For example, if
there is a "kernel+rootfs" partition, a single image can
be created which appends the rootfs after the kernel with
an appropriate search string. geom_map can be told to
search for said search string and create a partition
beginning after it.
If building (custom) FreeBSD images people tend to patch param.h. In case
this happens just before the build is started (within the same second)
CHECK_TIME actually triggers thinking param.h is in the future (see f_Xtime,
c_Xtime logi in find(1) sources for the details in !F_EXACTTIME case).
Using the -mtime -0s (seconds, rather than no unit) avoids this 1s race.
Submitted by: ed (2009-05-03)
Reviewed by: cperciva (2009-05-03), emaste
Tested by: bz (for almost two years)
MFC after: 4 days
Rick Macklem [Mon, 11 Apr 2011 20:54:30 +0000 (20:54 +0000)]
Vrele ni_startdir in the experimental NFS server for the case
of NFSv2 getting an error return from VOP_MKNOD(). Without this
patch, the server file system remains busy after an NFSv2
VOP_MKNOD() fails.
Add an install target to release/Makefile instead of grubbing around
(and requesting that users grub around) in /usr/obj. Also make sure that
make clean removes the bootonly media and clean up that target a little.
Alexander Motin [Mon, 11 Apr 2011 08:23:27 +0000 (08:23 +0000)]
Rework change made at r203146. Instead of reporting all wire errors as
SCSI status errors to CAM (that was wrong, as it too often turned retriable
wire errors into non-retriable REQUEST SENSE errors), do it only for STALL
errors on control pipe of the CBI devices. STALL on control pipe is just
a one of the ways to report error for CBI devices.
Jeff Roberson [Mon, 11 Apr 2011 01:43:59 +0000 (01:43 +0000)]
- Refactor softdep_setup_freeblocks() into a set of functions to prepare
for a new journal specific partial truncate routine.
- Use dep_current[] in place of specific dependency counts. This is
automatically maintained when workitems are allocated and has
less risk of becoming incorrect.
Rick Macklem [Sun, 10 Apr 2011 20:43:07 +0000 (20:43 +0000)]
Add some cleanup code to the module unload operation for
the experimental NFS server, so that it doesn't leak memory
when unloaded. However, unloading the NFSv4 server is not
recommended, since all NFSv4 state will be lost by the unload
and clients will have to recover the state after a server
reload/restart as if the server crashed/rebooted.
Rick Macklem [Sun, 10 Apr 2011 14:11:07 +0000 (14:11 +0000)]
Fix a bug in the userland rpc library, where it would use a
negative return value from write to update its position in
a buffer. The patch, courtesy of Andrey Simonenko, also simplifies
a conditional by removing the "i != cnt" clause, since it is
always true at this point in the code. The bug caused problems
for mountd, when it generated a large reply to an exports RPC
request.
Submitted by: simon at comsys.ntu-kpi.kiev.ua
MFC after: 2 weeks
Jeff Roberson [Sun, 10 Apr 2011 03:49:53 +0000 (03:49 +0000)]
Fix a long standing SUJ performance problem:
- Keep a hash of indirect blocks that have recently been freed and are
still referenced in the journal.
- Lookup blocks in this hash before forcing a new block write to wait on
the journal entry to hit the disk. This is only necessary to avoid
confusion between old identities as indirects and new identities as
file blocks.
- Don't free jseg structures until the journal has written a record that
invalidates it. This keeps the indirect block information around for
as long as is required to be safe.
- Force an empty journal block write when required to flush out stale
journal data that is simply waiting for the oldest valid sequence
number to advance beyond it.
Rick Macklem [Sun, 10 Apr 2011 01:54:42 +0000 (01:54 +0000)]
Add support for a backup stable restart file to the nfsd,
used for NFSv4 restart. This permits the nfsd to create
the stable restart file as required and minimizes the risk
of trouble if the file is lost.
Rick Macklem [Sat, 9 Apr 2011 23:55:27 +0000 (23:55 +0000)]
Add a VOP_UNLOCK() for the directory, when that is not what
VOP_LOOKUP() returned. This fixes a bug in the experimental
NFS server for the case where VFS_VGET() fails returning EOPNOTSUPP
in the ReaddirPlus RPC, forcing the use of VOP_LOOKUP() instead.
Linuxolator calls VOP_READDIR with ncookies pointer. Implement a
workaround for fdescfs to not panic when ncookies is not NULL, similar
to the one committed as r152254, but simpler, due to fdescfs_readdir()
not calling vfs_read_dirent().
Rename the "Reboot" option to "Exit", which reflects what it actually
does. The rc.local wrapper script will also soon grow an option to use the
live environment after installation, so an immediate reboot isn't even
necessarily implied.
Remove a check in udp6_send() that prevented v4-mapped v6 addresses from
working. We store v4 and v6 addresses as a union but for v4-mapped
addresses only store the 32bits w/o the ::ffff: word. That failed the
check as for example 127.0.0.1 would be ::7f00:1 rather than ::ffff:7f00:1
and the IN6_IS_ADDR_V4MAPPED() never worked here. Given we can hardly get
here with an unbound local address or invalid inp_vflags remove the check.
Reported by: tuexen
Reviewed by: tuexen
MFC after: 3 days
After r219579 and r219779 unbreak v4-mapped v6 sockets for UDP
some more. Similar to what we do for TCP check for v4-mapped
addresses and then handle them or the normal v6 address case.
For either set inp_vflags before calling into the pcb connect
function so that we have an unambiguous view in case we need to
set the local address or port.
Looked at: tuexen (as part of more)
MFC after: 3 days
Remove setting of PCB_FULL_IRET at the places where we are going to call
update_gdt_{f,g}sbase. The functions set the flag when td == curthread,
and sysarch is always called with curthread.
Disable local interrupts before testing the PCB_FULL_IRET flag.
Thread might be preempted after testing, which causes the flag to be
cleared. If ast was not delivered, we will do sysret with potentially
wrong fs/gs bases.
Reintroduce the fix already discussed in r216805 (please check its history
for a detailed explanation of the problems).
The only difference with the previous fix is in Solution2:
CPUBLOCK is no longer set when exiting from callout_reset_*() functions,
which avoid the deadlock (leading to r217161).
There is no need to CPUBLOCK there because the running-and-migrating
assumption is strong enough to avoid problems there.
Furthermore add a better !SMP compliancy (leading to shrinked code and
structures) and facility macros/functions.
Alexander Motin [Fri, 8 Apr 2011 14:42:29 +0000 (14:42 +0000)]
- Add kern.cam.ada.X.write_cache tunables/sysctls to control write caching
on per-device basis.
- While adding support for per-device sysctls, merge from graid branch
support for ADA_TEST_FAILURE kernel option, which opens few more sysctl,
allowing to simulate read and write errors for testing purposes.
John Baldwin [Fri, 8 Apr 2011 13:33:57 +0000 (13:33 +0000)]
Fix a bug in the previous change to restore the fast path for syscall
return. The ast() function may cause a context switch in which case
PCB_FULL_IRET would be set in the pcb. However, the code was not
rechecking the flag after ast() returned and would not properly restore
the FSBASE and GSBASE MSRs. To fix, recheck the PCB_FULL_IRET flag after
ast() returns.
While here, trim an instruction (and memory access) from the doreti path
and fix a typo in a comment.
When specifying the -t option (send tag in front of message), this tag
should also be forwarded to the remote logging host, not only when the
logging is done locally.
Adrian Chadd [Fri, 8 Apr 2011 09:20:45 +0000 (09:20 +0000)]
Add initial support for MIMO statistics to net80211.
This introduces struct ieee80211_rx_stats - which stores the various kinds
of RX statistics which a MIMO and non-MIMO 802.11 device can export.
It also fleshes out the mimo export to userland (node_getmimoinfo()).
It assumes that MIMO radios (for now) export both ctl and ext channels.
Non-11n MIMO radios are possible (and I believe Atheros made at least
one), so if that chipset support is added, extra flags to the
struct ieee80211_rx_stats can be added to extend this support.
Two new input functions have been added - ieee80211_input_mimo() and
ieee80211_input_mimo_all() - which MIMO-aware devices can call with
MIMO specific statistics.
802.11 devices calling the non-MIMO input functions will still function.
Adrian Chadd [Fri, 8 Apr 2011 07:44:00 +0000 (07:44 +0000)]
Begin fleshing out a public HAL routine to export the per-chain
ctl/ext noise floor values.
This routine doesn't check to see whether the radio is MIMO
capable - instead, it simply returns either the raw values,
the "nominal" values if the raw values aren't yet available
or are invalid, or '0' values if there's no valid channel/
no valid MIMO values.
Callers are expected to verify the radio is a MIMO radio
(which for now means it's an 11n chipset, there are non-11n
MIMO chipsets out there but I don't think we support them,
at least in MIMO mode) before exporting the MIMO values.
Adrian Chadd [Fri, 8 Apr 2011 06:58:01 +0000 (06:58 +0000)]
Export the per-chain ctl/ext noise floor values, raw and uncut, to the
upper-level HAL.
Right now the per-chain noise floor values aren't used anywhere in
the upper-level HAL, so the driver currently has no real reference
to compare the per-chain RSSI values to.
This is needed before per-chain RSSI values (for ctl and ext radios)
are can be thrown upstairs to the net80211 code.
Jung-uk Kim [Thu, 7 Apr 2011 23:28:28 +0000 (23:28 +0000)]
Use atomic load & store for TSC frequency. It may be overkill for amd64 but
safer for i386 because it can be easily over 4 GHz now. More worse, it can
be easily changed by user with 'machdep.tsc_freq' tunable (directly) or
cpufreq(4) (indirectly). Note it is intentionally not used in performance
critical paths to avoid performance regression (but we should, in theory).
Alternatively, we may add "virtual TSC" with lower frequency if maximum
frequency overflows 32 bits (and ignore possible incoherency as we do now).
John Baldwin [Thu, 7 Apr 2011 21:32:25 +0000 (21:32 +0000)]
If a system call does not request a full interrupt return, use a fast
path via the sysretq instruction to return from the system call. This was
removed in 190620 and not quite fully restored in 195486. This resolves
most of the performance regression in system call microbenchmarks between
7 and 8 on amd64.
- Fixed a problem when multiqueue mode was enabled which caused the CQ
chain to be corrupted.
- Removed many console print warnings and replaced with driver maintained
counters.
- Several style(9) fixes.
- Adjust a comment to actual behaviour
- Makefile nit
- Add more CVS/SVN keywords to make it easier to track changes from NetBSD
in case they add further improvements
Approved by: delphij (mentor)
Obtained from: The NetBSD Project