emaste [Wed, 7 May 2014 00:51:24 +0000 (00:51 +0000)]
Merge -fstandalone-debug from Clang r198655:
Implement a new -fstandalone-debug option. rdar://problem/15685848
It controls everything that -flimit-debug-info used to, plus the
vtable type optimization. The old -fno-limit-debug-info option is now an
alias to -fstandalone-debug and vice versa.
Standalone is the default on Darwin until dtrace is updated to work with
non-standalone debug info (rdar://problem/15758808).
Note: I kept the LimitedDebugInfo name in CodeGenOptions::DebugInfoKind
because NoStandaloneDebugInfo sounded even more confusing.
asomers [Tue, 6 May 2014 22:06:39 +0000 (22:06 +0000)]
dd(1) uses gettimeofday(2) to compute the throughput statistics. However,
gettimeofday returns the system clock, which may jump forward or back,
especially if NTP is in use. If the time jumps backwards, then dd will see
negative elapsed time, round it up to 1usec, and print an absurdly fast
transfer rate.
The solution is to use clock_gettime(2) with CLOCK_MONOTONIC_PRECISE as the
clock_id. That clock advances steadily, regardless of changes to the system
clock.
marcel [Tue, 6 May 2014 21:54:05 +0000 (21:54 +0000)]
Add mkimg_write() which combines lseek(2) and write(2) and uses
sector granularity for both offset and length. Have all schemes
use mkimg_write() instead of mkimg_seek() followed by write(2).
Now that schemes don't use lseek(2) nor write(2) directly, it's
easier to support output formats other than raw disks.
scottl [Tue, 6 May 2014 20:40:16 +0000 (20:40 +0000)]
Due to reasons unknown at this time, the system can be forced to write
a journal block even when there are no journal entries to be written.
Until the root cause is found, handle this case by ensuring that a
valid journal segment is always written.
Second, the data buffer used for writing journal entries was never
being scrubbed of old data. Fix this.
Submitted by: Takehara Mikihito
Obtained from: Netflix, Inc.
MFC after: 3 days
delphij [Tue, 6 May 2014 19:03:04 +0000 (19:03 +0000)]
Import George Wilson's change for Illumos #4730:
4730 metaslab group taskq should be destroyed in metaslab_group_destroy()
Reviewed by: Alex Reece <alex.reece@delphix.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Sebastien Roy <sebastien.roy@delphix.com>
markj [Tue, 6 May 2014 18:07:58 +0000 (18:07 +0000)]
Add a postinit debugger hook to rtld. This will be used by dtrace(1) to halt
the victim process before its entry point is called, at which point probes
and DOF data are registered with the kernel. The r_debug_state hook cannot
be used for this purpose, as it is called before the program's init routines
are invoked and in particular before DOF data is registered (via drti.o).
tuexen [Tue, 6 May 2014 16:51:07 +0000 (16:51 +0000)]
Remove unused code. This is triggered by the bugreport of Sylvestre Ledru
which deal with useless code in the user land stack:
https://bugzilla.mozilla.org/show_bug.cgi?id=1003929
marius [Tue, 6 May 2014 16:29:02 +0000 (16:29 +0000)]
- Allow foot shooting with the resetconfig command via the -f option.
- Fix typos preventing -f to actually work with the create command.
- Initialize flags to zero rather than using stack garbage when handling
the grow command.
MFC after: 3 days
Sponsored by: Bally Wulff Games & Entertainment GmbH
imp [Tue, 6 May 2014 15:44:46 +0000 (15:44 +0000)]
No need to install man pages for bootstrapping new make. Some
environments (that I can't reproduce locally, but that others have
reported) seem to get tripped up by this man page install. There's
really no need to do it, so turn off the man pages using the most
portable method. We can't just directly set MK_MAN=no here because
we're bootstrapping in the host environment and such a setting was
forbidden until very recently. NO_MAN= can produce a warning, but for
now the warning is benign.
imp [Tue, 6 May 2014 14:33:18 +0000 (14:33 +0000)]
Fix the upgrade path for fmake by allowing the include of src.opts.mk
to fail and falling back on the conservative testing of variables
already in place. This should stop the tide of tinderbox mail.
ian [Tue, 6 May 2014 14:03:35 +0000 (14:03 +0000)]
Call platform_pl310_init() before enabling the controller, and handle the
case where the controller is already enabled.
Some of the pl310 configuration registers cannot be changed while the
controller is active, so if there is any platform-specific init to be done
it must happen before enabling the controller.
The controller should not be enabled upon entry to the kernel, but u-boot
has recently developed the bad habit of leaving caches enabled when
launching the kernel, and since we have no control over that source code
we have to do our best to cope with it. The PL310 manual doesn't document
a safe sequence for disabling the controller, but the sequence used here
(force write-through mode and disable linefill allocations, then clean and
invalidate the current contents before disabling the hardware) appears to
be sound both by analysis and empirical testing.
These changes were developed and tested in collaboration with
Svatopluk Kraus <onwahe@gmail.com>.
Fix include statement to accept src.opts.mk from a source directory
instead of from /usr/share/mk.
I'm not sure that this will let buildworld complete on a system with
no installed src.opts.mk (make buildworld is still running), but the
tinderbox builds are all failing earlyon without this patch.
hselasky [Tue, 6 May 2014 09:12:32 +0000 (09:12 +0000)]
Reduce the number of interrupts in USB host mode for the DWC OTG
controller driver by piggybacking the SOF interrupt when issuing new
and checking old transfers. Number of interrupts was reduced by 30%
when doing Isochronous transfers.
Use correct GINTMSK_XXX macros when accessing the DWC OTG interrupt
mask register.
Add code to adjust the frame interval register which influences the
SOF rate.
ken [Tue, 6 May 2014 06:18:43 +0000 (06:18 +0000)]
Fix a problem with async notifications in the mpr(4) driver.
This problem only occurs on versions of FreeBSD prior to the recent CAM
locking changes. (i.e. stable/9 and older versions of stable/10) This
change should be a no-op for head and stable/10.
If a path isn't specified, xpt_register_async() will create a fully
wildcarded path and acquire a lock (the XPT lock in older versions,
and via xpt_path_lock() in newer versions) to call xpt_action() for the
XPT_SASYNC_CB CCB. It will then drop the lock and if the requested event
includes AC_FOUND_DEVICE or AC_PATH_REGISTERED, it will get the caller up
to date with any device arrivals or path registrations.
The issue is that before the locking changes, each SIM lock would get
acquired in turn during the EDT tree traversal process. If a path is
specified for xpt_register_async(), it won't acquire and drop its own lock,
but instead expects the caller to hold its own SIM lock. That works for
the first part of xpt_register_async(), but causes a recursive lock
acquisition once the EDT traversal happens and it comes to the SIM in
question. And it isn't possible to call xpt_action() without holding a SIM
lock.
The locking changes fix this by using the XPT topology lock for EDT
traversal, so it is no longer an issue to hold the SIM lock while calling
xpt_register_async().
The solution for FreeBSD versions before the locking changes is to request
notification of all device arrivals (so we pass a NULL path into
xpt_register_async()) and then filter out the arrivals that are not ours.
MFC After: 3 days
Sponsored by: Spectra Logic Corporation
imp [Tue, 6 May 2014 04:22:37 +0000 (04:22 +0000)]
Remove support for WITHOUT_BMAKE. bmake is now the only make that can
build world, so it is the only make we build or install. fmake is
still in the tree, but disconnected, and upgrades from older systems
that still have bmake has not been removed, but its state has not been
tested (it should work given how minimal the work to upgrade to bmake
is).
imp [Tue, 6 May 2014 04:21:48 +0000 (04:21 +0000)]
Move the /usr/src specific options to src.opts.mk. Move inclusion of
/etc/src.conf to this file as well. Now, it will only affect builds of
/usr/src and not others that use the bsd.*.mk files. Specifically
don't install src.opts.mk so we can catch when it 'leaks' into
bsd.*.mk again and have there be errors when this happens. Future
commits will move to including src.opts.mk instead of bsd.own.mk when
all that's needed is one of the MK_FOO options from src.opts.mk.
Future options should be placed here, unless they directly affect a
bsd.*.mk file, in which case they should be placed in bsd.opts.mk.
alc [Tue, 6 May 2014 03:42:04 +0000 (03:42 +0000)]
Prior to r254304, a separate function, vm_pageout_page_stats(), was used to
periodically update the reference status of the active pages. This function
was called, instead of vm_pageout_scan(), when memory was not scarce. The
objective was to provide up to date reference status for active pages in
case memory did become scarce and active pages needed to be deactivated.
The active page queue scan performed by vm_pageout_page_stats() was
virtually identical to that performed by vm_pageout_scan(), and so r254304
eliminated vm_pageout_page_stats(). Instead, vm_pageout_scan() is
called with the parameter "pass" set to zero. The intention was that when
pass is zero, vm_pageout_scan() would only scan the active queue. However,
the variable page_shortage can still be greater than zero when memory is not
scarce and vm_pageout_scan() is called with pass equal to zero.
Consequently, the inactive queue may be scanned and dirty pages laundered
even though that was not intended by r254304. This revision fixes that.
adrian [Tue, 6 May 2014 01:15:42 +0000 (01:15 +0000)]
Modify the RX path to keep the previous RX descriptor around once it's
used.
It turns out that the RX DMA engine does the same last-descriptor-link-
pointer-re-reading trick that the TX DMA engine. That is, the hardware
re-reads the link pointer before it moves onto the next descriptor.
Thus we can't free a descriptor before we move on; it's possible the
hardware will need to re-read the link pointer before we overwrite
it with a new one.
Tested:
* AR5416, STA mode
TODO:
* more thorough AP and STA mode testing!
* test on other pre-AR9380 NICs, just to be sure.
* Break out the RX descriptor grabbing bits from the RX completion
bits, like what is done in the RX EDMA code, so ..
* .. the RX lock can be held during ath_rx_proc(), but not across
packet input.
glebius [Tue, 6 May 2014 00:00:07 +0000 (00:00 +0000)]
- Remove net.inet.tcp.reass.overflows sysctl. It counts exactly
same events that tcpstat's tcps_rcvmemdrop counter counts.
- Rename tcps_rcvmemdrop to tcps_rcvreassfull and improve its
description in netstat(1) output.
neel [Mon, 5 May 2014 23:54:13 +0000 (23:54 +0000)]
Disable the 'uart_drain()' callback when the emulated receive FIFO is full.
Failing to do this will cause the kevent(2) notification to trigger
continuously and the bhyve(8) mevent thread will hog the cpu until the
characters on the backend tty device are drained.
Also, make the uart backend file descriptor non-blocking to avoid a
select(2) before every byte read from that backend.
gjb [Mon, 5 May 2014 23:01:01 +0000 (23:01 +0000)]
Add a helper script for parsing 'svn log' for the
'Relnotes:' tag in case 'yes' is not explicitly
the first string value following the tab.
As it turns out, a number of commits have bypassed
my filters (both email and 'svn log --search'), and
this script returns the results I want when doing
these searches.
imp [Mon, 5 May 2014 22:03:08 +0000 (22:03 +0000)]
Simplify use of the semi-bogus EARLY_BUILD variable. Also, simplify
use of semi-bogus NO_WARNS variable. Both of these, in this case,
should collapse to WITHOUT_WARNS in a future commit.
imp [Mon, 5 May 2014 22:03:01 +0000 (22:03 +0000)]
Introduce CWARNFLAGS.${COMPILER_TYPE} and use that to simplify the
thicket of .if ${COMPILER_TYPE} == "clang" that controls
warnings. Also, use CFLAGS.clang in a couple places in preference to
having a similar construct that's related to the CWARNFLAGS changes.
imp [Mon, 5 May 2014 22:02:48 +0000 (22:02 +0000)]
[1] Make WITHOUT_FOO alway trump WITH_FOO, regardless of the system
default. This restores more of the historical expectations that
were broken when we started disallowing both WITH_FOO and
WITHOUT_FOO to be defined.
[2] Document this new behavior, and improve the documentation in
general here.
ray [Mon, 5 May 2014 21:44:53 +0000 (21:44 +0000)]
Set of updates to vt(4) core part.
o Declare vt(4) drivers dataset.
o Create single static structures for all early drivers.
o Add vt(4) to be by default in the kernel consoles list.
o Create one more sysinit point, to be able to initialize memory and lock
requirement of early drivers.
o Implement early drivers select. (Only best available will be selected).
o Fix one missed "return (0)" for VTYLOCK.
o Improve locking for cases when one driver replace another.
o Make driver replacement notification less debug-look-like.
ken [Mon, 5 May 2014 19:53:03 +0000 (19:53 +0000)]
Adjust #if statements inside mprsas_send_smpcmd() to more accurately
reflect when unmapped I/O support was added.
For FreeBSD 10, it arrived just prior to __FreeBSD_version 1000028.
For FreeBSD 9, it arrived just prior to __FreeBSD_version 902001.
Also, fix compiler warnings in mprsas_send_smpcmd() that happen in the
i386 PAE build for non-unmapped I/O builds. These were fixed in mps(4)
in revision 241145, but didn't make it into the mpr(4) driver. This
change should only affect FreeBSD versions outside the above revisions,
and thus doesn't affect head.
MFC after: 3 days
Sponsored by: Spectra Logic Corporation
asomers [Mon, 5 May 2014 19:38:29 +0000 (19:38 +0000)]
Remove the ifconfig test added in rev 263445. After discussion with
melifaro, we agreed that ifconfig's behavior was not a bug. The main
motivation for bin/187551 was to partially resolve kern/187549, but we
resolved kern/187549 in a different way instead.
tuexen [Mon, 5 May 2014 17:30:05 +0000 (17:30 +0000)]
Fill in csum_data only for UDP or TCP packets. This fixes a bug where
SCTP were reported to have always a correct checksum if they don't
contain any ethernet padding.
neel [Mon, 5 May 2014 16:30:03 +0000 (16:30 +0000)]
Re-adding an event to a kqueue modifies the parameters of the original event.
However, if the original knote had been disabled then it is not automatically
re-enabled.
Fix this by using EV_ADD to create an mevent and EV_ENABLE to enable it.
Adding a kevent for the first time implicitly enables it so existing callers
of mevent_add() don't need to change.
hselasky [Mon, 5 May 2014 14:31:34 +0000 (14:31 +0000)]
Build the kernel sound module without ISA DMA support for ARM and MIPS
platforms, because these platforms do not implement the ISA DMA
API. Else the sound modules cannot be loaded when running these
platforms.
hselasky [Mon, 5 May 2014 11:50:52 +0000 (11:50 +0000)]
Improve DWC OTG USB host side support for isochronous FULL and HIGH
speed data traffic going directly to a USB device or through a
so-called USB transaction translator.
trasz [Mon, 5 May 2014 09:20:30 +0000 (09:20 +0000)]
Make r242379 - the fix for UFS labels disappearing after resizing
the provider - also apply to UFS1 filesystems. This should help with
resizing filesystems created by makefs(8), which still uses UFS1.
Tested by: jmg@
Sponsored by: The FreeBSD Foundation
adrian [Mon, 5 May 2014 08:12:21 +0000 (08:12 +0000)]
Break out the multicast programming into its own hardware specific
call, which assumes the hardware is awake.
Turn ath_update_mcast() into a routine that's only called from the
net80211 layer - and it forces the hardware awake first.
This fixes a LOR from the EDMA RX path which calls ath_mode_init()
with the RX lock held - the driver lock can't also be grabbed.
This path assumes that the ath_mode_init() callers all wake up
the NIC first.
glebius [Sun, 4 May 2014 23:25:32 +0000 (23:25 +0000)]
The FreeBSD-SA-14:08.tcp was a lesson on not doing acrobatics with
mixing on stack memory and UMA memory in one linked list.
Thus, rewrite TCP reassembly code in terms of memory usage. The
algorithm remains unchanged.
We actually do not need extra memory to build a reassembly queue.
Arriving mbufs are always packet header mbufs. So we got the length
of data as pkthdr.len. We got m_nextpkt for linkage. And we need
only one pointer to point at the tcphdr, use PH_loc for that.
In tcpcb the t_segq fields becomes mbuf pointer. The t_segqlen
field now counts not packets, but bytes in the queue. This gives
us more precision when comparing to socket buffer limits.
nwhitehorn [Sun, 4 May 2014 16:38:21 +0000 (16:38 +0000)]
Disable ACPI and P4TCC throttling by default, following discussion on
freebsd-current. These CPU speed control techniques are usually unhelpful
at best. For now, continue building the relevant code into GENERIC so that
it can trivially be re-enabled at runtime if anyone wants it.
smh [Sun, 4 May 2014 14:05:14 +0000 (14:05 +0000)]
Use a zio flag to prevent recursion of vdev_queue_io_done which can
cause stack overflow for IO's which return ZIO_PIPELINE_CONTINUE
from the zio_vdev_io_start stage and hence don't suspend and complete
in a different thread.
This prevents double fault panic on slow machines running ZFS on
GELI volumes which return EOPNOTSUPP directly to BIO_DELETE requests.
melifaro [Sat, 3 May 2014 16:38:05 +0000 (16:38 +0000)]
Remove additional fib checks from rtalloc1_fib.
It looks like current consumers are either unaware
of MRT (and uses RT_DEFAULT_FIB implicitly) or
know what thay are doing, In latter case they
will be either hit by KASSERT or ESCRH will be returned
due to NULL rnh.