Refine r280308. Do not completely disable timestamping of devfs nodes
on reads or writes, the time marks are used to display idle time by
w(1) [1]. Instead, use vfs.devfs.dotimes as the selector of default
precision vs. using time_second. The later gives seconds precision,
which is good enough for the purpose.
Note that timestamp updates are unlocked and the updates itself, as
well as the check in devfs_timestamp, are non-atomic.
Noted by: truckman [1]
Reviewed by: bde
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
adrian [Wed, 1 Apr 2015 03:42:46 +0000 (03:42 +0000)]
Start the process of migrating the ANI statistics out of the HALs and into
the top-level HAL.
The athstats program is blindly using a copy of the ar5212 ANI stats structure
to pull out ANI statistics/state and this is problematic for the AR9300
HAL.
So:
* Define HAL_ANI_STATS and HAL_ANI_STATE
* Use HAL_ANI_STATS inside the AR5212 HAL
This commit doesn't (yet) convert the ar5212AniState -> HAL_ANI_STATE when
exporting it to userland; that'll come in the next commit.
Summary:
Add "GELI Passphrase:" prompt to boot loader.
A new loader.conf(5) option of geom_eli_passphrase_prompt="YES" will now
allow you to enter your geli(8) root-mount credentials prior to invoking
the kernel.
Test Plan:
Drop a head copy of check-password.4th into /boot and then apply the patch
(only the patch to /boot/check-password.4th is required; no other changes are
required but you do have to have a HEAD copy of check-password.4th to
apply the patch).
NB: The rest of your /boot files can be up to 2 years old but no older.
NB: The test won't work unless your kernel has the following change
You should be prompted for a GELI passphrase before the menu (if enabled),
just after loading loader.conf(5).
NB: It doesn't matter if you're using GELI or not. However if you are using
GELI and a sufficiently new enough release (has SVN r273489) and you entered
the proper passphrase to mount your GELI encrypted root device(s), you should
notice that the boot process did not stop (you went from loader all the way to login).
Solve dreaded "dictionary full" issue by segregating words into new
vocabularies delay-processing, password-processing, version-processing,
frame-drawing, menu-infrastructure, menu-namespace, menu-command-helpers,
and menusets-infrastructure. The net effect is to remove almost 200
definitions from the main forth vocabulary reducing the dictionary size
by over 50%. The chances of hitting "dictionary full" should be greatly
reduced by this patch.
Fix overflow bugs in and remove obsolete limit from kernel RPC
implementation.
The kernel RPC code, which is responsible for the low-level scheduling
of incoming NFS requests, contains a throttling mechanism that
prevents too much kernel memory from being tied up by NFS requests
that are being serviced. When the throttle is engaged, the RPC layer
stops servicing incoming NFS sockets, resulting ultimately in
backpressure on the clients (if they're using TCP). However, this is
a very heavy-handed mechanism as it prevents all clients from making
any requests, regardless of how heavy or light they are. (Thus, when
engaged, the throttle often prevents clients from even mounting the
filesystem.) The throttle mechanism applies specifically to requests
that have been received by the RPC layer (from a TCP or UDP socket)
and are queued waiting to be serviced by one of the nfsd threads; it
does not limit the amount of backlog in the socket buffers.
The original implementation limited the total bytes of queued requests
to the minimum of a quarter of (nmbclusters * MCLBYTES) and 45 MiB.
The former limit seems reasonable, since requests queued in the socket
buffers and replies being constructed to the requests in progress will
all require some amount of network memory, but the 45 MiB limit is
plainly ridiculous for modern memory sizes: when running 256 service
threads on a busy server, 45 MiB would result in just a single
maximum-sized NFS3PROC_WRITE queued per thread before throttling.
Removing this limit exposed integer-overflow bugs in the original
computation, and related bugs in the routines that actually account
for the amount of traffic enqueued for service threads. The old
implementation also attempted to reduce accounting overhead by
batching updates until each queue is fully drained, but this is prone
to livelock, resulting in repeated accumulate-throttle-drain cycles on
a busy server. Various data types are changed to long or unsigned
long; explicit 64-bit types are not used due to the unavailability of
64-bit atomics on many 32-bit platforms, but those platforms also
cannot support nmbclusters large enough to cause overflow.
This code (in a 10.1 kernel) is presently running on production NFS
servers at CSAIL.
Summary of this revision:
* Removes 45 MiB limit on requests queued for nfsd service threads
* Fixes integer-overflow and signedness bugs
* Avoids unnecessary throttling by not deferring accounting for
completed requests
Fix "MOVS" instruction memory to MMIO emulation. Currently updates to
%rdi, %rsi, etc are inadvertently bypassed along with the check to
see if the instruction needs to be repeated per the 'rep' prefix.
Add "MOVS" instruction support for the 'MMIO to MMIO' case.
jilles [Tue, 31 Mar 2015 20:51:01 +0000 (20:51 +0000)]
wordexp: Explicitly pass along IFS.
Per Austin group issue #884, sh should not import IFS from the environment
but always set it to $' \t\n'. For wordexp(), however, it is documented and
useful for it to use IFS from the environment.
Since sh currently imports IFS from the environment, this change has no
functional effect.
emaste [Tue, 31 Mar 2015 13:37:32 +0000 (13:37 +0000)]
unwind-d2 build workaround for arm64
The __builtin_init_dwarf_reg_size_table function is unimplemented in
clang 3.6 for AArch64. Comment it out for now and replace it with
a message and abort.
Tracked in upstream LLVM PR 22997
https://llvm.org/bugs/show_bug.cgi?id=22997
ganbold [Tue, 31 Mar 2015 11:50:46 +0000 (11:50 +0000)]
Add necessary changes to support various Amlogic SoC devices
specially aml8726-m6 and aml8726-m8b SoC based devices.
aml8726-m6 SoC exist in devices such as Visson ATV-102.
Hardkernel ODROID-C1 board has aml8726-m8b SoC.
The following support is included:
Basic machdep code
SMP
Interrupt controller
Clock control driver (aka gate)
Pinctrl
Timer
Real time clock
UART
GPIO
I2C
SD controller
SDXC controller
USB
Watchdog
Random number generator
PLL / Clock frequency measurement
Frame buffer
Submitted by: John Wehle
Approved by: stas (mentor)
ngie [Tue, 31 Mar 2015 06:43:55 +0000 (06:43 +0000)]
Cleanup and do minor refactoring before converting testcases to ATF
- Convert errx(-1, ..) to errx(1, ..)
- Move the aio(4) checks to a single function (aio_available); use modfind(2)
instead of depending on SIGSYS (doesn't work when aio(4) support is missing,
not documented in the aio syscall manpages).
- Use aio_available liberally in the testcase functions
- Use mkstemp(3) + unlink(2) instead of mktemp(3)
- Fix some -Wunused warnings
- Bump WARNS to 6
cperciva [Tue, 31 Mar 2015 01:44:29 +0000 (01:44 +0000)]
Partially revert r278118 now that the required logic for deciding whether
freebsd-update can be useful has moved into the firstboot_freebsd_update
script.
kib [Tue, 31 Mar 2015 01:44:07 +0000 (01:44 +0000)]
Provide workaround for a performance issue with the popcnt instruction
on Intel processors. Clear spurious dependency by explicitely xoring
the destination register of popcnt.
Use bitcount64() instead of re-implementing SWAR locally, for
processors without popcnt instruction.
Reviewed by: jhb
Discussed with: jilles (previous version)
Sponsored by: The FreeBSD Foundation
np [Tue, 31 Mar 2015 01:22:20 +0000 (01:22 +0000)]
cxgbe/tom: return rx credits promptly if the socket buffer's low water
mark cannot be reached because the window advertised to the peer isn't
wide enough. While here, tweak the normal credit return too.
jhb [Mon, 30 Mar 2015 20:13:22 +0000 (20:13 +0000)]
Wait 100 microseconds for a local APIC to dispatch each startup-related IPI
rather than 20. The MP 1.4 specification states in Appendix B.2:
"A period of 20 microseconds should be sufficient for IPI dispatch to
complete under normal operating conditions".
(Note that this appears to be separate from the 10 millisecond (INIT) and
200 microsecond (STARTUP) waits after the IPIs are dispatched.) The
Intel SDM is silent on this issue as far as I can tell.
At least some hardware requires 60 microseconds as noted in the PR, so
bump this to 100 to be on the safe side.
emaste [Mon, 30 Mar 2015 20:01:41 +0000 (20:01 +0000)]
llvm: Backport upstream r229195 to fix arm64 TLS relocations
As is described at http://llvm.org/bugs/show_bug.cgi?id=22408, the GNU
linkers ld.bfd and ld.gold currently only support a subset of the
whole range of AArch64 ELF TLS relocations. Furthermore, they assume
that some of the code sequences to access thread-local variables are
produced in a very specific sequence. When the sequence is not as the
linker expects, it can silently mis-relaxe/mis-optimize the
instructions.
Even if that wouldn't be the case, it's good to produce the exact
sequence, as that ensures that linkers can perform optimizing
relaxations.
This patch:
* implements support for 16MiB TLS area size instead of 4GiB TLS area
size. Ideally clang would grow an -mtls-size option to allow support
for both, but that's not part of this patch.
* by default doesn't produce local dynamic access patterns, as even
modern ld.bfd and ld.gold linkers do not support the associated
relocations. An option (-aarch64-elf-ldtls-generation) is added to
enable generation of local dynamic code sequence, but is off by
default.
* makes sure that the exact expected code sequence for local dynamic
and general dynamic accesses is produced, by making use of a new
pseudo instruction. The patch also removes two
(AArch64ISD::TLSDESC_BLR, AArch64ISD::TLSDESC_CALL) pre-existing
AArch64-specific pseudo SDNode instructions that are superseded by
the new one (TLSDESC_CALLSEQ).
dim [Mon, 30 Mar 2015 19:36:33 +0000 (19:36 +0000)]
Pull in r233552 from upstream libc++ trunk (by Eric Fiselier):
[libcxx] Fix PR22771 - Support access control SFINAE in the library
version of is_convertible.
Summary:
Currently the conversion check does not take place in a context where
access control SFINAE is applied. This patch changes the context of
the test expression so that SFINAE occurs if access control does not
permit the conversion.
Related bug: https://llvm.org/bugs/show_bug.cgi?id=22771
emaste [Mon, 30 Mar 2015 17:38:25 +0000 (17:38 +0000)]
Switch to ELF toolchain readelf(1)
ELF toolchain readelf lacked some functionality at the time other tools
(like size, strip, nm, etc.) were switched over to the ELF toolchain
versions. That has been addressed as of the last update, so we can add
it to the list.
PR: 198950 [exp-run]
Reviewed by: bapt, imp, rpaulo
Relnotes: yes
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D2156
mav [Mon, 30 Mar 2015 13:30:53 +0000 (13:30 +0000)]
Periodically wake up threads waiting for vmem(9) resources, so they could
ask for resource reclamation again.
This is kind of dirty hack, but as last resort this is better then stuck
indefinitely because of KVA fragmentation, waiting until some random event
free something sufficient. OpenSolaris also has this hack in its vmem(9).
zbb [Mon, 30 Mar 2015 09:49:54 +0000 (09:49 +0000)]
Fix bug in xrefinfo_find() for 64-bit platforms
uintptr_t may be 64-bit on some platforms, therefore when
finding xrefinfo by pointer to device the high word is being
cut off due to cast to phandle_t which is 32-bit long by definition.
Due to that we loose the high word of the address to compare with
xi->dev's address.
To fix that, first argument of xrefinfo_find() is extended to
uintptr_t and is being cast to appropriate type (phandle_t)
when compared.
cperciva [Mon, 30 Mar 2015 08:33:19 +0000 (08:33 +0000)]
Clean up filesystem unmounting in vmimage builds:
* Remove vm_umount_base function which is currently unused.
* Add umount_loop function which loops attempting to unmount one filesystem.
* Replace calls to umount with calls to umount_loop.
* Don't attempt to unmount ${DESTDIR}/dev if it isn't mounted.
The looping is necessary because sometimes umount fails due to filesystems
being busy. The most common cause of such busyness is periodic(8) jobs
running `find / ...`.
royger [Mon, 30 Mar 2015 07:09:07 +0000 (07:09 +0000)]
xen: add a handler for the debug interrupt
Handle the VIRQ_DEBUG signal and print a stack trace of each vCPU on the Xen
console. This is only used for debug purposes and is triggered by the
administrator of the Xen host.
Sponsored by: Citrix Systems R&D
MFC after: 1 week
markj [Mon, 30 Mar 2015 04:06:36 +0000 (04:06 +0000)]
Fix ping(8) and ping6(8) usage in a couple of ip provider tests, and
update expected test output to reflect differences in default TTL and
payload length.
markj [Mon, 30 Mar 2015 04:01:49 +0000 (04:01 +0000)]
Replace dtest.pl, the upstream DTrace test suite harness, with a shell
script. This reimplementation is much simpler than dtest.pl and is more
amenable to being run under Kyua - dtest.pl writes error output to a
temporary directory that is deleted when the run finishes, making it hard
to debug test failures. This change also removes the test suite's dependency
on perl.
This adds an upper bound, dtrace_ustackdepth_max, to the number of frames
traversed when computing the userland stack depth. Some programs - notably
firefox - are otherwise able to trigger an infinite loop in
dtrace_getustack_common(), causing a panic.
adrian [Sun, 29 Mar 2015 21:53:08 +0000 (21:53 +0000)]
Add initial support for the HAL channel survey support to the AR9300 HAL.
This is used by the 'athsurvey' command to print out channel survey
statistics - % busy times transmit, receive and airtime.
It's as buggy and incomplete as the rest of the HAL survey support -
notably, tying into the ANI code to read channel stats and occasionally
getting garbage counters isn't very nice. It also doesn't (yet!) get
channel survey information during a scan. But it's good enough for
basic air-time debugging, which is why I'm committing it in this state.
adrian [Sun, 29 Mar 2015 21:50:21 +0000 (21:50 +0000)]
Move the HAL channel survey support out to be in the top-level HAL,
rathe than private in each HAL module.
Whilst here, modify ath_hal_private to always have the per-channel
noisefloor stats, rather than conditionally. This just makes
life easier in general (no strange ABI differences between different
HAL compile options.)
Add a couple of methods (clear/reset, add) rather than using
hand-rolled versions of things.