imp [Sun, 21 Dec 2014 05:07:11 +0000 (05:07 +0000)]
Where appropriate, use the modern terms for the one true time base
(UTC) rather than the archaic (GMT) in comments. Except where the
comments are making fun of people doing this (and pedants who insist
on the new terms).
adrian [Sun, 21 Dec 2014 04:58:45 +0000 (04:58 +0000)]
Update ieee80211_sta_tim_notify() to do double duty - handle STA sleep
to awake transition as well as handle waking up a VAP in STA powersave
mode if it's in bgscan.
This was a reasonably hairy bug to try and figure out and it became
more obvious because of stuff I've done.
Specifically:
* a NIC would go into bgscan mode - either because of a bgscan timer
or wpa_supplicant asked it to;
* the AP would indicate there's traffic for the STA by setting the TIM
bitmap bit for it;
* mindwell would be met during scan, so it'd wake up and break out of
the scan loop in scan_task(), but
* because the scan wasn't completed, it wouldn't bring the VAP out of
STA mode powersave (so it wouldn't tell the AP about it and it would
block VAP TX);
* .. but because we kept seeing the TIM bit set, ic->ic_lastdata was
being constantly updated, and ..
* bgscancont() would thus never say "yes we can continue a bgscan"
so the bgscan would hang and never make progress.
Now, I do see this particular state occur on iwn(4) - /however/ -
this NIC has the firmware call ieee80211_scan_next() once the firmware
scan for that channel has completed. This has the effect of moving
the scan along to the next channel. I do see the debug that I'm adding
where we see a beacon with a TIM bit set whilst we're in bgscan, so
the condition about waking up to receive traffic is triggering.
It just won't cause a hang.
For other NICs - all of the USB ones and at least ath(4) -
ieee80211_scan_next() / ieee80211_scan_done() isn't called.
So it relies upon the mindwell timer, the beacon receive and the
beacon / probe response -> ieee80211_add_scan() to move along
the scan state.
In the above case, mindwell triggered, there's no beacons triggering
the scan_add code to move things along, and we weren't waking things
up when seeing the TIM set for us. So it just hung until the interface
was dropped.
So, the short-term fix here is to do what the comment in scan_task()
says - if we are in bgscan mode and we see our TIM bit set, just wake
up the VAP. If it's already awake then it's a nop. If we're awake
then we transition to awake and handle the traffic. Once there's no
TX or RX traffic going on, ic->ic_lastdata won't be updated anymore
and bgscancont() will continue.
This was triggered more often after my initial SLEEP state handling
for software sleep states - because now I update ic->ic_lastdata
upon seeing a TIM bit set, not just the RX of the subsequent traffic.
That's needed so the thing doesn't ping-pong up and down between
seeing the TIM bit set, sending the "I'm awake" NULL data frame, and
starting to receive data from the AP.
I'd like to subsequently split ic_lastdata into two - one for TX and
one for RX - so it becomes easier to use the correct one (or both!)
when making decisions like whether to scan, go to sleep, etc.
I'd appreciate this getting some further testing.
Tested:
* rsu(4), STA mode, bgscan on
* iwn(4), STA mode, bgscan on
glebius [Sat, 20 Dec 2014 22:12:04 +0000 (22:12 +0000)]
Revert r274494, r274712, r275955 and provide extra comments explaining
why there could appear a zero-sized mbufs in socket buffers.
A proper fix would be to divorce record socket buffers and stream
socket buffers, and divorce pru_send that accepts normal data from
pru_send that accepts control data.
adrian [Sat, 20 Dec 2014 19:41:31 +0000 (19:41 +0000)]
Document where in scan_task the scan state can change, and potentially
deal/log a warning if the scan flags change during one of those race
windows.
It's highly likely that I need to actually sit down and replace this
scan infrastructure at some point. It has some other side effects too -
the scan task is a blocking task scheduled in the net80211 taskqueue;
so drivers that use this taskqueue have other things not run. Eek.
If you see this printf happen then please let me know!
glebius [Sat, 20 Dec 2014 14:19:46 +0000 (14:19 +0000)]
Add to sbappendstream_locked() a check against NULL mbuf, like it is done
in sbappend_locked() and sbappendrecord_locked().
This is a quick fix to the panic introduced by r274712.
A proper solution should be to make sosend_generic() avoid calling
pru_send() with NULL mbuf for the protocols that do not understand
control messages. Those protocols that understand control messages,
should be able to receive NULL mbuf, if control is non-NULL.
neel [Sat, 20 Dec 2014 04:57:45 +0000 (04:57 +0000)]
Various 8259 device model improvements:
- implement 8259 "polled" mode.
- set 'atpic->sfn' if bit 4 in ICW4 is set during master initialization.
- report error if guest tries to enable the "special mask" mode.
ian [Sat, 20 Dec 2014 04:24:40 +0000 (04:24 +0000)]
Log mmc and sd command failures. Reporting of routine expected errors,
such as timeouts while probing a bus or testing for a feature, is
squelched. Also, error reporting is limited to 5 events per second,
because when an sdcard goes bad on a low-end embedded board, flooding
the console at high speed isn't helpful.
Original logging code contributed by Michal Meloun, but then I fancied
it up with squelching and ppsratecheck.
ian [Sat, 20 Dec 2014 01:13:13 +0000 (01:13 +0000)]
Add a new sdhci quirk, SDHCI_QUIRK_WAITFOR_RESET_ASSERTED, to work around
TI OMAP controllers which will return the reset-in-progress bit as zero if
you read the status register too fast after setting the reset bit.
The zero is apparently from a stale snapshot of the internal state presented
in the interface register, and leads to a false indication that the reset
is complete when it either hasn't started yet or is in-progress. The
workaround is to first loop until the bit is seen as asserted, then do the
normal loop waiting to see it de-asserted.
imp [Sat, 20 Dec 2014 00:04:01 +0000 (00:04 +0000)]
Remove support for FreeBSD 7 and really old FreeBSD 8. The classifiers
have been in the base for a while, so the gymnastics here aren't
needed. In addition, the bugs in subr_disk.c have been fixed since
2009, so there's no need for an identical copy of it in the tree
anymore. There's really no need to binary patch g_io_request, so let's
get rid of the code (not compiled in anymore) lest others think it is
a good idea.
benno [Fri, 19 Dec 2014 19:09:22 +0000 (19:09 +0000)]
Adjust the test of a KASSERT to better match the intent.
This assertion was added in r246213 as a guard against corrupted mbufs
arriving from drivers, the key distinguishing factor of said mbufs being
that they had a negative length. Given we're in a while loop specifically
designed to skip over zero-length mbufs, panicking on a zero-length mbuf
seems incorrect.
imp [Fri, 19 Dec 2014 18:45:52 +0000 (18:45 +0000)]
Bump the largest record we can cope with from 1k to 8k. Other users of
the hints file don't have any real limits, and longer records will
need to be written shortly.
delphij [Thu, 18 Dec 2014 23:45:26 +0000 (23:45 +0000)]
MFV r275914:
As of r270383, the dbuf_compare comparator compares the dbuf
attributes in the following order:
db_level (indirect level)
db_blkid (block number)
db_state (current state)
the address of the element
Because db_state is being considered before the element's state,
changing of db_state would affect balancedness of the AVL tree,
even when the address of element compares differently. For
instance, in dbuf_create, db_state may be altered after the
node is inserted into the AVL tree and may break AVL tree
balancedness.
Instead of using db_state as a comparision critera (introduced
in r270383), consider it only when we are doing a lookup, that
is one of the two dbuf pointers contains DB_SEARCH.
Illumos issue:
5422 preserve AVL invariants in dn_dbufs
ngie [Thu, 18 Dec 2014 18:16:00 +0000 (18:16 +0000)]
Fix building/installing tests when TESTSBASE != /usr/tests
The work in r258233 hardcoded the assumption that tests was the last component
of the tests tree by pushing tests as an explicit prefix for the paths in
BSD.tests.dist and /usr was the prefix for all tests, per BSD.usr.dist and all
of the mtree calls used in Makefile.inc1. This assumption breaks if/when one
provides a custom TESTSBASE "prefix", e.g. TESTSBASE=/mytests .
One thing that r258233 did properly though was remove "/usr/tests" creation
from BSD.usr.dist -- that should have not been there in the first place. That
was an "oops" on my part for the work that was originally committed in r241823
jamie [Thu, 18 Dec 2014 18:10:39 +0000 (18:10 +0000)]
Setgid before running a command as a specified user. Previously only
initgroups(3) was called, what isn't quite enough. This brings jail(8)
in line with jexec(8), which was already doing the right thing.
imp [Thu, 18 Dec 2014 16:57:22 +0000 (16:57 +0000)]
Don't deselect the card too soon. To set the block size or switch the
function parameters, the card has to be in transfer state. If it is in
the idle state, the commands are ignored. This caused us not to set
the proper parameters that we later assume to be present, leading to
downstream failures of the card / interface as our state machine
mismatches the card's.
Submitted by: Svatopluk Kraus <onwahe at gmail.com>, Michal Meloun
<meloun at miracle.cz>
kib [Thu, 18 Dec 2014 10:01:12 +0000 (10:01 +0000)]
The VOP_LOOKUP() implementations for CREATE op do not put the name
into namecache, to avoid cache trashing when doing large operations.
E.g., tar archive extraction is not usually followed by access to many
of the files created.
Right now, each VOP_LOOKUP() implementation explicitely knowns about
this quirk and tests for both MAKEENTRY flag presence and op != CREATE
to make the call to cache_enter(). Centralize the handling of the
quirk into VFS, by deciding to cache only by MAKEENTRY flag in VOP.
VFS now sets NOCACHE flag for CREATE namei() calls.
Note that the change in semantic is backward-compatible and could be
merged to the stable branch, and is compatible with non-changed
third-party filesystems which correctly handle MAKEENTRY.
Suggested by: Chris Torek <torek@pi-coral.com>
Reviewed by: mckusick
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
adrian [Thu, 18 Dec 2014 05:17:18 +0000 (05:17 +0000)]
Fix the scan handling for 11b->11g upgrades in a world where, well,
it's not just 11b/11g.
The following was happening, and it's quite .. annoyingly grr-y.
* create vap, setup wpa_supplicant with no bgscanning, etc - there's
no call to ieee80211_media_change, so vap->iv_des_mode is
IEEE80211_MODE_AUTO;
* do ifconfig wlan0 scan - same thing, media_change doesn't get called,
iv_des_mode stays as auto.
* But then, run wpa_cli and do 'scan' - it'll do a media change.
* if you're on 11ng, vap->iv_des_mode gets changed to IEEE80211_MODE_11NG
* Then makescanlist() is called. There's a block of code that gets
called if iv_des_mode != IEEE80211_MODE_AUTO, and it does this:
* .. now, iv_des_mode is not IEEE80211_MODE_11G, so it always runs
'continue'
* .. and thus the scan list stays empty and no further channel
scans occur. Ever.(1)
If you then disassociate and try associating to something, your
scan table has likely been purged / aged out and you'll never
see anything in the scan list.
(1) You need to do 'ifconfig wlan0 mode auto' or just destroy/re-create
the VAP to get working wireless again.
Tested:
* iwn(4) - intel 5300 wifi; STA mode; using wpa_supplicant; bgscan
enabled -and- wpa_supplicant scanning.
Thanks to:
* Everyone who kept poking me about this and wondering why the hell
their wifi would eventually stop seeing scan lists. Grr.
I eventually snapped this evening and dug back into this code.
dteske [Thu, 18 Dec 2014 03:51:09 +0000 (03:51 +0000)]
In bsdinstall's distextract, replace mixed_gauge() of dialog(3) with
new dpv(3) wrapper to dialog(3) dialog_gauge(). The dpv(3) library provides
a more flexible and refined interface similar to dialog_mixedgauge() however
is implemented atop the more generalized dialog_gauge() for portability.
Noticeable improvements in bsdinstall's distextract will be a status line
showing data rate information (with support for localeconv(3) to format
numbers according to $LANG or $LC_ALL conversion information), i18n support,
improved auto-sizing of gauge widget, a ``wheel barrow'' to keep the user
informed that things are moving (even if status/progress has not changed),
improved color support (mini-progress bars use the same color, if enabled,
as the main gauge bar), and several other improvements (some not visible).
dpv stands for "dialog progress view" (dpv was introduced in SVN r274116).
Differential Revision: https://reviews.freebsd.org/D714
Discussed on: -current
Reviewed by: julian
MFC after: 3 days
X-MFC-to: stable/10
Relnotes: Improved installer feedback from bsdinstall distextract
ngie [Wed, 17 Dec 2014 20:02:07 +0000 (20:02 +0000)]
Fix sporadic build failures due to race when running make installworld
when strip gets replaced at install time by adding it to ITOOLS for the
default usr.bin/xinstall STRIP_CMD
This will fix the failure noted in this Jenkins build step:
https://jenkins.freebsd.org/job/Build-UFS-image/688/
This will also fix the issue reported by alfred@ dealing with installing on
targets that differ from build hosts (e.g. installing on i386/i386 when built
on amd64/amd64)
mav [Wed, 17 Dec 2014 17:30:54 +0000 (17:30 +0000)]
Add configuration options to override physical and UNMAP blocks geometry.
While in most cases CTL should correctly fetch those values from backing
storages, there are some initiators (like MS SQL), that may not like large
physical block sizes, even if they are true. For such cases allow override
fetched values with supported ones (like 4K).
mav [Wed, 17 Dec 2014 15:13:21 +0000 (15:13 +0000)]
Make sequence numbers checks more strict.
While we don't support MCS, hole in received sequence numbers may mean
only PDU loss. While we don't support lost PDU recovery, terminate the
connection to avoid stuck commands.
While there, improve handling of sequence numbers wrap after 2^32 PDUs.
emaste [Wed, 17 Dec 2014 14:46:21 +0000 (14:46 +0000)]
Do not strip all when stripping an explicit symbol
When requested to strip specific symbols (-N flag) the default should be
to strip nothing (other than the requested symbols). This is consistent
with binutils strip(1).
PR: 196038
Reviewed by: imp
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D1327
will [Wed, 17 Dec 2014 00:22:41 +0000 (00:22 +0000)]
Initialize an argument to NULL instead of expecting dlinfo() to do it.
dlinfo() is a weak reference that may not be initialized at the time of
execution. The default implementation (in lib/libc/gen/dlfcn.c) neither
modifies the address pointed to by the third argument nor returns an error.
kib [Tue, 16 Dec 2014 18:28:33 +0000 (18:28 +0000)]
The iret instruction may generate #np and #ss fault, besides #gp.
When returning to usermode, the handler for that exceptions is also
executed with wrong gs base. Handle all three possible faults in the
same way, checking for iret fault, and performing full iret.
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
will [Tue, 16 Dec 2014 17:59:05 +0000 (17:59 +0000)]
Make NanoBSD source-able from other scripts.
Summary:
This change converts NanoBSD into a two-script bundle.
- defaults.sh contains all non-CLI code. Most NanoBSD code is moved into
this file.
- nanobsd.sh now consists just of a command line interface that calls into
functions in defaults.sh.
Test Plan: Run NanoBSD using a previously-working configuration.
ae [Tue, 16 Dec 2014 14:59:20 +0000 (14:59 +0000)]
Add ability to not specify a zone identifier twice, when both source and
destination addresses are specified.
For example:
# ping6 -S fe80::1%ix0 ff02::1
or
# ping6 -S fe80::1 fe80::2%ix0
ed [Tue, 16 Dec 2014 09:21:56 +0000 (09:21 +0000)]
Rename cpack*() to CMPLX*().
The C11 standard introduced a set of macros (CMPLX, CMPLXF, CMPLXL) that
can be used to construct complex numbers from a pair of real and
imaginary numbers. Unfortunately, they require some compiler support,
which is why we only define them for Clang and GCC>=4.7.
The cpack() function in libm performs the same task as CMPLX(), but
cannot be used to generate compile-time constants. This means that all
invocations of cpack() can safely be replaced by C11's CMPLX(). To keep
the code building with GCC 4.2, provide copies of CMPLX() that can at
least be used to generate run-time complex numbers.
This makes it easier to build some of the functions outside of libm.
neel [Tue, 16 Dec 2014 06:33:57 +0000 (06:33 +0000)]
For level triggered interrupts clear the PIC IRR bit when the interrupt pin
is deasserted. Prior to this change each assertion on a level triggered irq
pin resulted in two interrupts being delivered to the CPU.
yongari [Tue, 16 Dec 2014 06:13:30 +0000 (06:13 +0000)]
Fix a bug introdiced in r217548. According to NS DP83815 data
sheet, RX filter should be disabled before programming.
Previously it was clearing wrong bits so RX filter was not
disabled in RX filter configuration.
delphij [Mon, 15 Dec 2014 18:28:22 +0000 (18:28 +0000)]
MFV r275784:
Plug a memory leak in libzfs. In zfs_iter_bookmarks, an nvlist is allocated
before calling lzc_get_bookmarks, which allocates the nvlist again (and
overwrites the pointer to previously allocated list).
Illumos issue:
5427 memory leak in libzfs when doing rollback
delphij [Mon, 15 Dec 2014 18:22:45 +0000 (18:22 +0000)]
MFV r275783:
Convert ARC flags to use enum. Previously, public flags are defined in
arc.h and private flags are defined in arc.c which can lead to confusion
and programming errors.
Consistently use 'hdr' (when referencing arc_buf_hdr_t) instead of 'buf'
or 'ab' because arc_buf_t are often named 'buf' as well.
Illumos issue:
5369 arc flags should be an enum
5370 consistent arc_buf_hdr_t naming scheme
Calculate the segment's memory size (p_memsz) using the virtual
addresses, not the file offsets. Otherwise padding preceeding SHT_NOBITS
sections may be excluded from the calculation, resulting in a segment
that is too small.