freebsd.git
4 years agoposix_spawn: mark error as volatile
kevans [Tue, 21 Jan 2020 22:02:53 +0000 (22:02 +0000)]
posix_spawn: mark error as volatile

In the case of an error, the RFSPAWN'd thread will write back to psa->error
with the correct exit code. Mark this as volatile as the return value is
being actively dorked up for erroneous exits on !x86.

This fixes the following tests, tested on aarch64 (only under qemu, at the
moment):

- posix_spawn/spawn_test:t_spawn_missing
- posix_spawn/spawn_test:t_spawn_nonexec
- posix_spawn/spawn_test:t_spawn_zero

Reported by: mikael
MFC after: 3 days

4 years agoRevert r356948; breaks build somehow.
trasz [Tue, 21 Jan 2020 20:32:49 +0000 (20:32 +0000)]
Revert r356948; breaks build somehow.

4 years agoFix build when WITHOUT_WPA_SUPPLICANT_EAPOL option used.
cy [Tue, 21 Jan 2020 20:21:52 +0000 (20:21 +0000)]
Fix build when WITHOUT_WPA_SUPPLICANT_EAPOL option used.

The build failure was discoved by Michael Dexter's recent Build Options
Survey run, at https://callfortesting.org/results/bos-2020-01-16/\
WITHOUT_WPA_SUPPLICANT_EAPOL-small.txt.

Reported by: Michael Dexter <editor@callfortesting.org> via emaste
MFC after: 2 weeks

4 years agoMake linux(4) handle MAP_32BIT.
trasz [Tue, 21 Jan 2020 19:19:02 +0000 (19:19 +0000)]
Make linux(4) handle MAP_32BIT.

This unbreaks Mono (mono-devel-4.6.2.7+dfsg-1ubuntu1 from Ubuntu Bionic);
previously would crash on "amd64_is_imm32" assert.

MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation

4 years agoAdd relocation handling required for -zifunc-noplt to work on arm64.
markj [Tue, 21 Jan 2020 17:45:49 +0000 (17:45 +0000)]
Add relocation handling required for -zifunc-noplt to work on arm64.

Static relocations for the immediate operand of a branch instruction
must be applied.

In a patch which implements LSE-based atomic(9) operations using ifuncs,
-zifunc-noplt reduces system CPU usage during a buildkernel by several
percent.

Also fix elf_reloc_internal() to return an error if symbol lookup fails.

Reviewed by: andrew
MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D17392

4 years agoRegen.
markj [Tue, 21 Jan 2020 17:28:36 +0000 (17:28 +0000)]
Regen.

MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation

4 years agoFix 64-bit syscall argument fetching in 32-bit Linux syscall handlers.
markj [Tue, 21 Jan 2020 17:28:22 +0000 (17:28 +0000)]
Fix 64-bit syscall argument fetching in 32-bit Linux syscall handlers.

The Linux32 system call argument fetcher places each argument (passed in
registers in the Linux x86 system call convention) into an entry in the
generic system call args array.  Each member of this array is 8 bytes
wide, so this approach is broken for system calls that take off_t
arguments.

Fix the problem by splitting l_loff_t arguments in the 32-bit system
call descriptions, the same as we do for FreeBSD32.  Change entry points
to handle this using the PAIR32TO64 macro.

Move linux_ftruncate64() into compat/linux.

PR: 243155
Reported by: Alex S <iwtcex@gmail.com>
Reviewed by: kib (previous version)
MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D23210

4 years agolibexec/rc/rc.d/ipfw: style fix after r356943
eugen [Tue, 21 Jan 2020 17:16:02 +0000 (17:16 +0000)]
libexec/rc/rc.d/ipfw: style fix after r356943

Also, make sure it does not break for systems without ipfw code loaded.

MFC after: 1 months
X-MFC-with: 356943

4 years agoCorrect "service ipfw status" for INET6-only systems.
eugen [Tue, 21 Jan 2020 17:02:57 +0000 (17:02 +0000)]
Correct "service ipfw status" for INET6-only systems.

MFC after: 1 month

4 years agoCorrect a misleading indent.
brooks [Tue, 21 Jan 2020 16:31:08 +0000 (16:31 +0000)]
Correct a misleading indent.

This dates to before the beginning of our repo and was found clang 10.

MFC after: 3 days
Sponsored by: DARPA

4 years agoAdd support for Hygon Dhyana Family 18h processor.
kib [Tue, 21 Jan 2020 13:22:35 +0000 (13:22 +0000)]
Add support for Hygon Dhyana Family 18h processor.

As a new x86 CPU vendor, Chengdu Haiguang IC Design Co., Ltd (Hygon)
is a joint venture between AMD and Haiguang Information Technology Co.,
Ltd., aims at providing x86 processors for China server market.

The first generation Hygon processor(Dhyana) shares most architecture
with AMD's family 17h, but with different CPU vendor ID("HygonGenuine")
and PCI vendor ID(0x1d94) and family series number 18h(Hygon negotiated
with AMD to confirm that only Hygon use family 18h).

To enable Hygon Dhyana support in FreeBSD, add new definitions
HYGON_VENDOR_ID("HygonGenuine") and X86_VENDOR_HYGON(0x1d94) to identify
Hygon Dhyana CPU.

Initialize the CPU features(topology, local APIC ext, MSI, TSC, hwpstate,
MCA, DEBUG_CTL, etc) for amd64 and i386 mode by sharing the code path of
AMD family 17h.

The changes have been applied on FreeBSD 13.0-CURRENT and tested
successfully on Hygon Dhyana processor.

References:
[1] Linux kernel patches for Hygon Dhyana, merged in 4.20:

https://git.kernel.org/tip/c9661c1e80b609cd038db7c908e061f0535804ef

[2] MSR and CPUID definition:

https://www.amd.com/system/files/TechDocs/54945_PPR_Family_17h_Models_00h-0Fh.pdf

Submitted by: Pu Wen <puwen@hygon.cn>
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D23163

4 years agoDocument requirements for the 'struct route' variations.
melifaro [Tue, 21 Jan 2020 12:00:34 +0000 (12:00 +0000)]
Document requirements for the 'struct route' variations.

MFC after: 2 weeks

4 years agoMake sure the VNET is properly set when calling tcp_drop() from
hselasky [Tue, 21 Jan 2020 11:43:25 +0000 (11:43 +0000)]
Make sure the VNET is properly set when calling tcp_drop() from
the ktls taskqueue callback function.

A valid VNET is needed when updating statistics.

panic()
tcp_state_change()
tcp_drop()
ktls_reset_send_tag()
taskqueue_run_locked()
taskqueue_thread_loop()

Sponsored by: Mellanox Technologies

4 years agosysent.mk: split interpreter out of target command
kevans [Tue, 21 Jan 2020 05:01:11 +0000 (05:01 +0000)]
sysent.mk: split interpreter out of target command

The main objective here is to make it easy to identify what needs to change
in order to use a different sysent generator than the current Lua-based one,
which may be used to MFC some of the changes that have happened so we can
avoid parallel accidents in stable branches, for instance.

As a secondary objective, it's now feasible to override the generator on a
per-Makefile basis if needed, so that one could refactor their Makefile to
use this while pinning generation to the legacy makesyscalls.sh. I don't
anticipate any consistent need for such a thing, but it's low-effort to
achieve.

4 years agoMove readahead and dropbehind fault functionality into a helper routine for
jeff [Tue, 21 Jan 2020 00:12:57 +0000 (00:12 +0000)]
Move readahead and dropbehind fault functionality into a helper routine for
clarity.

Reviewed by: dougm, kib, markj
Differential Revision: https://reviews.freebsd.org/D23282

4 years agorandom.3: Some minor improvements to wording/clarity
cem [Mon, 20 Jan 2020 23:44:10 +0000 (23:44 +0000)]
random.3: Some minor improvements to wording/clarity

4 years agolibc: Delete unused rand.c ifdef TEST code
cem [Mon, 20 Jan 2020 23:43:47 +0000 (23:43 +0000)]
libc: Delete unused rand.c ifdef TEST code

4 years agoReduce object locking in vm_fault. Once we have an exclusively busied page we
jeff [Mon, 20 Jan 2020 22:49:52 +0000 (22:49 +0000)]
Reduce object locking in vm_fault.  Once we have an exclusively busied page we
no longer need an object lock.  This reduces the longest hold times and
eliminates some trylock code blocks.

Reviewed by: kib, markj
Differential Revision: https://reviews.freebsd.org/D23034

4 years agoMerge commit bc4bc5aa0 from llvm git (by Justin Hibbits):
dim [Mon, 20 Jan 2020 20:10:28 +0000 (20:10 +0000)]
Merge commit bc4bc5aa0 from llvm git (by Justin Hibbits):

  Add 8548 CPU definition and attributes

  8548 CPU is GCC's name for the e500v2, so accept this in clang.  The
  e500v2 doesn't support lwsync, so define __NO_LWSYNC__ for this as
  well, as GCC does.

  Differential Revision:  https://reviews.llvm.org/D67787

Merge commit ff0311c4b from llvm git (by Justin Hibbits):

  [PowerPC]: Add powerpcspe target triple subarch component

  Summary:
  This allows the use of '-target powerpcspe-unknown-linux-gnu' or
  'powerpcspe-unknown-freebsd' to be used, instead of '-target
  powerpc-unknown-linux-gnu -mspe'.

  Reviewed By: dim
  Differential Revision: https://reviews.llvm.org/D72014

Merge commit ba91dffaf from llvm git (by Fangrui Song):

  [Driver][PowerPC] Move powerpcspe logic from cc1 to Driver

  Follow-up of D72014. It is more appropriate to use a target feature
  instead of a SubTypeArch to express the difference.

  Reviewed By: #powerpc, jhibbits

  Differential Revision: https://reviews.llvm.org/D72433

commit 36eedfcb3 from llvm git (by Justin Hibbits):

  [PowerPC] Fix powerpcspe subtarget enablement in llvm backend

  Summary:

  As currently written, -target powerpcspe will enable SPE regardless
  of disabling the feature later on in the command line.  Instead,
  change this to just set a default CPU to 'e500' instead of a generic
  CPU.

  As part of this, add FeatureSPE to the e500 definition.

  Reviewed By: MaskRay
  Differential Revision: https://reviews.llvm.org/D72673

These are needed to unbreak the build for powerpcspe.

Requested by: jhibbits
MFC after: 1 week

4 years agoWhen MK_CASPER=no is set remove files which are not needed to run system.
oshogbo [Mon, 20 Jan 2020 19:56:22 +0000 (19:56 +0000)]
When MK_CASPER=no is set remove files which are not needed to run system.

PR: 242971

4 years agocache: revert r352613 now that vhold does not take locks
mjg [Mon, 20 Jan 2020 19:52:23 +0000 (19:52 +0000)]
cache: revert r352613 now that vhold does not take locks

4 years agoEven when the MK_CASPER is set to "no" we still want to install man pages
oshogbo [Mon, 20 Jan 2020 19:51:53 +0000 (19:51 +0000)]
Even when the MK_CASPER is set to "no" we still want to install man pages
and the headers. If the user decides to install the system without Casper
support, then the Casper functions are mocked, but they still exist in
the system.

PR: 242971
MFC after: 2 weeks

4 years agoThose files are already removed in ObsoleteFiles.\
oshogbo [Mon, 20 Jan 2020 19:47:58 +0000 (19:47 +0000)]
Those files are already removed in ObsoleteFiles.\
There is no need to remove them twice.

PR: 242971
MFC after: 2 weeks

4 years agoAdd myself (mikael) as a ports committer
mikael [Mon, 20 Jan 2020 19:38:29 +0000 (19:38 +0000)]
Add myself (mikael) as a ports committer

Approved by: manu (mentor)
Differential Revision: https://reviews.freebsd.org/D23227

4 years agoqsort.3: Bump Dd and note that Annex K is optional
cem [Mon, 20 Jan 2020 18:43:10 +0000 (18:43 +0000)]
qsort.3: Bump Dd and note that Annex K is optional

4 years agox86: Wait for curthread to be set up as an indicator that the boot stack
kib [Mon, 20 Jan 2020 17:23:03 +0000 (17:23 +0000)]
x86: Wait for curthread to be set up as an indicator that the boot stack
is no longer used.

pc_curthread is set by cpu_switch after it stopped using the old
thread (or boot) stack.  This makes the smp_after_idle_runnable()
function not dependent on the internals of the scheduler operations.

Reviewed by: markj
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D23276

4 years agoThe ports tree now accepts /usr/local/share/man as a directory for manpage
bapt [Mon, 20 Jan 2020 16:59:39 +0000 (16:59 +0000)]
The ports tree now accepts /usr/local/share/man as a directory for manpage
and will slowly transition from /usr/local/man to it. To reflect this remove
the documentation of the manpages being an exception in the layout of /usr/local

Reported by: Dan Nelson <dnelson_1901@yahoo.com> (via IRC)
MFC after: 3 days

4 years agoremove unused WITHOUT_PC_SYSINSTALL description
emaste [Mon, 20 Jan 2020 15:19:56 +0000 (15:19 +0000)]
remove unused WITHOUT_PC_SYSINSTALL description

pc-sysinstall was moved from the base system to ports in r351781.

Submitted by: driesm.michiels gmail com
Differential Revision: https://reviews.freebsd.org/D21647

4 years agocache: make numcachehv use counter(9) on all archs
mjg [Mon, 20 Jan 2020 14:42:11 +0000 (14:42 +0000)]
cache: make numcachehv use counter(9) on all archs

Requested by: kib

4 years agoAdd support for latest Intel I219 device, supported in Lenovo Carbon X1 v7
gnn [Mon, 20 Jan 2020 12:53:02 +0000 (12:53 +0000)]
Add support for latest Intel I219 device, supported in Lenovo Carbon X1 v7

MFC after: 2 weeks

4 years agoProperly translate MNT_FORCE flag to Linux umount2(2). Previously
trasz [Mon, 20 Jan 2020 12:16:32 +0000 (12:16 +0000)]
Properly translate MNT_FORCE flag to Linux umount2(2).  Previously
it worked by accident.

MFC after: 2 weeks
Sponsored by: DARPA

4 years agoAdd qsort_r(3) regression test.
trasz [Mon, 20 Jan 2020 11:45:18 +0000 (11:45 +0000)]
Add qsort_r(3) regression test.

MFC after: 2 weeks
Sponsored by: DARPA
Differential Revision: https://reviews.freebsd.org/D23206

4 years agoAdd qsort_s(3). Apart from the constraints, it also makes it easier
trasz [Mon, 20 Jan 2020 11:40:07 +0000 (11:40 +0000)]
Add qsort_s(3).  Apart from the constraints, it also makes it easier
to port software written for Linux variant of qsort_r(3).

Reviewed by: kib, arichardson
MFC after: 2 weeks
Relnotes: yes
Sponsored by: DARPA
Differential Revision: https://reviews.freebsd.org/D23174

4 years agopowerpc/amiga: Hide CPLD date and time printing behind bootverbose
jhibbits [Mon, 20 Jan 2020 04:01:35 +0000 (04:01 +0000)]
powerpc/amiga: Hide CPLD date and time printing behind bootverbose

There's no need to see the CPLD build date and time every boot.

4 years agoDon't hold the object lock while calling getpages.
jeff [Sun, 19 Jan 2020 23:47:32 +0000 (23:47 +0000)]
Don't hold the object lock while calling getpages.

The vnode pager does not want the object lock held.  Moving this out allows
further object lock scope reduction in callers.  While here add some missing
paging in progress calls and an assert.  The object handle is now protected
explicitly with pip.

Reviewed by: kib, markj
Differential Revision: https://reviews.freebsd.org/D23033

4 years agomps(4): add missing cam(4) dependency
cem [Sun, 19 Jan 2020 22:52:36 +0000 (22:52 +0000)]
mps(4): add missing cam(4) dependency

On a MINIMAL kernel, mps.ko wouldn't load because it uses the xpt_hold_boot
symbol from CAM, but didn't have a dependency on cam(4).

(CEM: Some context: when linking loaded modules, the kernel dynamic linker
only looks for definitions in explictly marked dependency modules.  Also,
the identical mpr(4) driver uses the same CAM function, but already had the
correct MODULE_DEPEND(), so no similar change is needed there.)

Submitted by: Greg V <greg AT unrelenting.technology>
Reviewed by: imp, myself
Differential Revision: https://reviews.freebsd.org/D23272

4 years agoCirrus-CI: add `make packages` to CI test
emaste [Sun, 19 Jan 2020 22:29:22 +0000 (22:29 +0000)]
Cirrus-CI: add `make packages` to CI test

Now that we can override the format to avoid the time spent compressing
pkgbase packages we can test `make packages` with only 5 minutes or so
added to the CI cycle time.

A future change should switch the CI smoke test to using these packages.

Sponsored by: The FreeBSD Foundation

4 years agoPowerPC: Add CPLD driver for AmigaOne X5000
jhibbits [Sun, 19 Jan 2020 21:43:15 +0000 (21:43 +0000)]
PowerPC: Add CPLD driver for AmigaOne X5000

Summary:
The CPLD is the communications medium between the CPU and the XMOS
"Xena" event coprocessor.  It provides a mailbox communication feature,
along with dual-port RAM to be used between the CPU and XMOS.  Also, it
provides basic board stats as well, such as PCIe presence, JTAG signals,
and CPU fan speed reporting (in revolutions per second).  Only fan speed
reading is handled, as a sysctl.

Reviewed by: bdragon
Differential Revision: https://reviews.freebsd.org/D23136

4 years agovfs: switch vop_stdunlock to call lockmgr_unlock
mjg [Sun, 19 Jan 2020 21:41:34 +0000 (21:41 +0000)]
vfs: switch vop_stdunlock to call lockmgr_unlock

Since the flags argument is now alawys 0 the new call provides the same
behavior.

4 years agox86: fix assertion in ipi_send_cpu to range check the passed id
mjg [Sun, 19 Jan 2020 21:35:51 +0000 (21:35 +0000)]
x86: fix assertion in ipi_send_cpu to range check the passed id

Prior to the change for sufficiently bad id (and in particular NOCPU which is -1)
it would access memory outside of the cpu_apic_ids array.

4 years ago[PowerPC64] fix crash when using machdep.moea64_bpvo_pool_size tunable
jhibbits [Sun, 19 Jan 2020 21:17:57 +0000 (21:17 +0000)]
[PowerPC64] fix crash when using machdep.moea64_bpvo_pool_size tunable

Summary:
This fixes kernel crashing when tunable "machdep.moea64_bpvo_pool_size" is
set to a value higher then 327680 (default value).  Function
moea64_mid_bootstrap() relies on moea64_bpvo_pool_size, but at time of the
use the variable wan't yet updated with the new value provided by user.

Problem was detected after trying to use a VM with 64GB of RAM, and default
moea64_bpvo_pool_size is insufficient (kernel boot used more than 470000) .
I think default value must be discussed to address this use case, or find a
way to calculate pool size automatically based on amount of memory detected.

Test Plan: Tested on QEMU VM with 64GB of RAM using "set
machdep.moea64_bpvo_pool_size=655360" on loader prompt

Submitted by: Alfredo Dal'Ava JĂșnior (alfredo.junior_eldorado.org.br)
Differential Revision: https://reviews.freebsd.org/D23233

4 years agozilinx/zy7_qspi: Add a qspi driver for Zynq platforms.
manu [Sun, 19 Jan 2020 20:04:44 +0000 (20:04 +0000)]
zilinx/zy7_qspi: Add a qspi driver for Zynq platforms.

This is a qspi driver for the Xilinx Zynq-7000 chip.
It could be useful for anyone wanting to boot a system from flash memory
instead of SD cards.

Submitted by: Thomas Skibo (thomasskibo@yahoo.com)
Differential Revision: https://reviews.freebsd.org/D14698

4 years agosrc.conf.5: regen after BINUTILS defaults and description changes
emaste [Sun, 19 Jan 2020 19:57:14 +0000 (19:57 +0000)]
src.conf.5: regen after BINUTILS defaults and description changes

4 years agork805: Add a regnode_init method
manu [Sun, 19 Jan 2020 19:56:50 +0000 (19:56 +0000)]
rk805: Add a regnode_init method

This method will set the desired voltaged based on values in the DTS.
It will not enable the regulator, this is the job of either a consumer
or regnode_set_constraint SYSINIT if the regulator is boot_on or always_on.

Reviewed by: mmel
MFC after: 1 month
Differential Revision: https://reviews.freebsd.org/D23216

4 years agosrc.opts.mk: default BINUTILS_BOOTSTRAP to NO except for x86 and powerpc
emaste [Sun, 19 Jan 2020 19:56:05 +0000 (19:56 +0000)]
src.opts.mk: default BINUTILS_BOOTSTRAP to NO except for x86 and powerpc

x86 needs bootstrap GNU as for assembling a few files, and powerpc needs
GNU ld.bfd for linking 32-bit objects.  All other targets either fully
use in-tree Clang and lld, or rely on external toolchain.

Sponsored by: The FreeBSD Foundation

4 years agoarm: allwinner: Add GPIO Interrupt support
manu [Sun, 19 Jan 2020 19:51:20 +0000 (19:51 +0000)]
arm: allwinner: Add GPIO Interrupt support

Not all pins in Allwinner have interrupts support so we rely
on the padconf data to add the proper caps when pin_getcaps is called.
The pin is switch to the specific "eint" function during setup_intr and
switched back to its old function in teardown_intr.
Only INTR_MAP_DATA_GPIO is supported for now.

MFC after: 1 month

4 years agoremove caution notes from WITHOUT_BINUTILS* descriptions
emaste [Sun, 19 Jan 2020 19:47:04 +0000 (19:47 +0000)]
remove caution notes from WITHOUT_BINUTILS* descriptions

WITHOUT_BINUTILS and WITHOUT_BINUTILS_BOOTSTRAP previously included
claims about being unable to build if set.  Those cautions are no longer
universally true, and most FreeBSD targets can function more or less
without enabling GNU Binutils.  Just remove the cautions.

Sponsored by: The FreeBSD Foundation

4 years agolimit building GNU assembler (as) to x86
emaste [Sun, 19 Jan 2020 19:16:32 +0000 (19:16 +0000)]
limit building GNU assembler (as) to x86

GNU as 2.17.50 is currently required by amd64 and i386 for at least one
file that cannot be assembled by Clang's integrated assembler (IAS).
Other supported CPU architectures either use Clang IAS for all assembly
files, or rely on external toolchain.

Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D23180

4 years agoarm: allwinner: Fix padconf for interrupts information
manu [Sun, 19 Jan 2020 19:14:49 +0000 (19:14 +0000)]
arm: allwinner: Fix padconf for interrupts information

Add a eint_bank member to the allwinner_pins structure.
On Allwinner SoCs not all pins can do interrupt.
Older SoC (A10/A13 and A20) there is a maximum number of interrupts
set to 32 and all the configuration is done in the same registers.
While on "newer" SoCs (>=A31) interrupts registers are splitted per
pin bank (i.e. all interrupts available in bank B will be configured
with a sets of registers and the one in bank G in another set).
While here set the names to all interrupts function to
pX_eintY where X is the bank name and Y the interrupt number.

To whom ever in the future look at the H5 manual and notice that the bank F
have interrupts support : This isn't true, trust me.

MFC after: 1 month

4 years agoIt has not been possible to recursively terminate a vnode object for some time
jeff [Sun, 19 Jan 2020 18:36:03 +0000 (18:36 +0000)]
It has not been possible to recursively terminate a vnode object for some time
now.  Eliminate the dead code that supports it.

Approved by: kib, markj
Differential Revision: https://reviews.freebsd.org/D22908

4 years agoMake collapse synchronization more explicit and allow it to complete during
jeff [Sun, 19 Jan 2020 18:30:23 +0000 (18:30 +0000)]
Make collapse synchronization more explicit and allow it to complete during
paging.

Shadow objects are marked with a COLLAPSING flag while they are collapsing with
their backing object.  This gives us an explicit test rather than overloading
paging-in-progress.  While split is on-going we mark an object with SPLIT.
These two operations will modify the swap tree so they must be serialized
and swap_pager_getpages() can now directly detect these conditions and page
more conservatively.

Callers to vm_object_collapse() now will reliably wait for a collapse to finish
so that the backing chain is as short as possible before other decisions are
made that may inflate the object chain.  For example, split, coalesce, etc.
It is now safe to run fault concurrently with collapse.  It is safe to increase
or decrease paging in progress with no lock so long as there is another valid
ref on increase.

This change makes collapse more reliable as a secondary benefit.  The primary
benefit is making it safe to drop the object lock much earlier in fault or
never acquire it at all.

This was tested with a new shadow chain test script that uncovered long
standing bugs and will be integrated with stress2.

Reviewed by: kib, markj
Differential Revision: https://reviews.freebsd.org/D22908

4 years agoProvide an API for interlocked refcount sleeps.
jeff [Sun, 19 Jan 2020 18:18:17 +0000 (18:18 +0000)]
Provide an API for interlocked refcount sleeps.

Reviewed by: kib, markj
Differential Revision: https://reviews.freebsd.org/D22908

4 years agovfs: allow v_holdcnt to transition 0->1 without the interlock
mjg [Sun, 19 Jan 2020 17:47:04 +0000 (17:47 +0000)]
vfs: allow v_holdcnt to transition 0->1 without the interlock

Since r356672 ("vfs: rework vnode list management") there is nothing to do
apart from altering freevnodes count, but this much can be safely done based
on the result of atomic_fetchadd.

Reviewed by: kib
Tested by: pho
Differential Revision: https://reviews.freebsd.org/D23186

4 years agocache: counter_u64_add_protected -> counter_u64_add
mjg [Sun, 19 Jan 2020 17:05:26 +0000 (17:05 +0000)]
cache: counter_u64_add_protected -> counter_u64_add

Fixes booting on RISC-V where it does happen to not be equivalent.

Reported by: lwhsu

4 years agoAdd myself as a mentee of bcr
carlavilla [Sun, 19 Jan 2020 16:24:25 +0000 (16:24 +0000)]
Add myself as a mentee of bcr

Patch by: carlavilla@(doc-committer)
Approved by: bcr@(mentor)

4 years agopkgbase: allow the pkg format to be overridden
emaste [Sun, 19 Jan 2020 14:46:28 +0000 (14:46 +0000)]
pkgbase: allow the pkg format to be overridden

Compressing .txz packages can be rather slow, and speed is likely more
important than disk space during development.  Allow package format to
be set via PKG_FORMAT make variable.

Reviewed by: bapt
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D23257

4 years agocache: convert numcachehv to counter(9) on 64-bit platforms
mjg [Sun, 19 Jan 2020 05:37:27 +0000 (05:37 +0000)]
cache: convert numcachehv to counter(9) on 64-bit platforms

4 years agovfs: plug a conditional assigment of lo_name in getnewvnode
mjg [Sun, 19 Jan 2020 05:36:45 +0000 (05:36 +0000)]
vfs: plug a conditional assigment of lo_name in getnewvnode

It only matters for witness. No functional changes.

4 years ago[PowerPC] Fix 32-bit ubldr calling convention
bdragon [Sun, 19 Jan 2020 04:13:19 +0000 (04:13 +0000)]
[PowerPC] Fix 32-bit ubldr calling convention

Due to the way u-boot for 32-bit powerpc is compiled, the interrupt code
assumes that the GOT pointer (r30) on u-boot is always intact.

When making syscalls to u-boot, ensure that we have restored r30 like we
found it before we enable interrupts to prevent u-boot from crashing if a
timer interrupt was pending.

This fixes ubldr on e500 qemu (assuming you have recompiled qemu's u-boot
with API support!)

Reviewed by: jhibbits
Sponsored by: Tag1 Consulting, Inc.
Differential Revision: https://reviews.freebsd.org/D23258

4 years agolibbe: use the new zfs_mount_at()
kevans [Sun, 19 Jan 2020 02:48:56 +0000 (02:48 +0000)]
libbe: use the new zfs_mount_at()

More background is available in r356876, but this new interface is more
portable across ZFS implementations and cleaner for what libbe is attempting
to achieve anyways.

MFC after: 3 days

4 years agolibzfs: add zfs_mount_at
kevans [Sun, 19 Jan 2020 02:45:02 +0000 (02:45 +0000)]
libzfs: add zfs_mount_at

This will be used in libbe in place of the internal zmount(); libbe only
wants to be able to mount a dataset at an arbitrary mountpoint without
altering dataset/pool properties. The natural way to do this in a portable
way is by creating a zfs_mount_at() interface that's effectively zfs_mount()
+ a mountpoint parameter. zfs_mount() is now a light wrapper around the new
method.

The interface and implementation have already been accepted into ZFS On
Linux, and the next commit to switch libbe() over to this new interface will
solve the last compatibility issue with ZoL.  The next sysutils/openzfs
rebase against ZoL should be able to build libbe/bectl with only minor
adjustments to build glue.

Reviewed by: Ryan Moeller <ryan freqlabs com>
MFC after: 3 days
Differential Revision: https://reviews.freebsd.org/D23132

4 years agopkgbase: cache pkg ABI for all world/kernel packages
emaste [Sat, 18 Jan 2020 22:58:32 +0000 (22:58 +0000)]
pkgbase: cache pkg ABI for all world/kernel packages

Rather than invoking `pkg config ABI` repeatedly.

Reviewed by: manu
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D23255

4 years agosysent targets: further cleanup and deduplication
kevans [Sat, 18 Jan 2020 20:37:45 +0000 (20:37 +0000)]
sysent targets: further cleanup and deduplication

r355473 vastly improved the readability and cleanliness of these Makefiles.
Every single one of them follows the same pattern and duplicates the exact
same logic.

Now that we have GENERATED/SRCS, split SRCS up into the two parameters we'll
use for ${MAKESYSCALLS} rather than assuming a specific ordering of SRCS and
include a common sysent.mk to handle the rest. This makes it less tedious to
make sweeping changes.

Some default values are provided for GENERATED/SYSENT_*; almost all of these
just use a 'syscalls.master' and 'syscalls.conf' in cwd, and they all use
effectively the same filenames with an arbitrary prefix. Most ABIs will be
able to get away with just setting GENERATED_PREFIX and including
^/sys/conf/sysent.mk, while others only need light additions. kern/Makefile
is the notable exception, as it doesn't take a SYSENT_CONF and the generated
files are spread out between ^/sys/kern and ^/sys/sys, but it otherwise fits
the pattern enough to use the common version.

Reviewed by: brooks, imp
Nice!: emaste
Differential Revision: https://reviews.freebsd.org/D23197

4 years agoCirrus-CI: bump VM image to FreeBSD 12.1
emaste [Sat, 18 Jan 2020 20:19:07 +0000 (20:19 +0000)]
Cirrus-CI: bump VM image to FreeBSD 12.1

12.0 will be EOL.

Sponsored by: The FreeBSD Foundation

4 years agopcpu_page_alloc: guard against empty NUMA domains
gallatin [Sat, 18 Jan 2020 18:25:37 +0000 (18:25 +0000)]
pcpu_page_alloc: guard against empty NUMA domains

Some systems, such as higher end Threadripper, may have
NUMA domains with no physical memory, Don't allocate
from these domains.

This fixes a "panic: vm_wait in early boot" on my 2990WX desktop

Reviewed by: jeff
Sponsored by: Netflix

4 years agoifa_maintain_loopback_route: adjust debugging output
eugen [Sat, 18 Jan 2020 04:48:05 +0000 (04:48 +0000)]
ifa_maintain_loopback_route: adjust debugging output

Correction after r333476:

- write this as LOG_DEBUG again instead of LOG_INFO;
- get back function name into the message;
- error may be ESRCH if an address is removed in process (by carp f.e.),
not only ENOENT;
- expression complexity grows, so try making it more readable.

MFC after: 1 week

4 years agoD23057: [PowerPC] Fix offset calculations in bridge mode
bdragon [Sat, 18 Jan 2020 04:12:41 +0000 (04:12 +0000)]
D23057: [PowerPC] Fix offset calculations in bridge mode

In rS354701, I replaced text relocations with offsets from &generictrap.

Unfortunately, the magic variable I was using doesn't actually mean the
address of &generictrap, in bridge mode it actually means &generictrap64.

So, for bridge mode to work, it is necessary to differentiate between
"where do we need to branch to to handle a trap" and "where is &generictrap
for purposes of doing relative math".

Introduce a new TRAP_ENTRY and use it instead of TRAP_GENTRAP for doing
actual calls to the generic trap handler.

Reported by: Mark Millard <marklmi@yahoo.com>
Reviewed by: jhibbits
Sponsored by: Tag1 Consulting, Inc.
Differential Revision: https://reviews.freebsd.org/D23057
> Description of fields to fill in above:                     76 columns --|
> PR:                       If and which Problem Report is related.
> Submitted by:             If someone else sent in the change.
> Reported by:              If someone else reported the issue.
> Reviewed by:              If someone else reviewed your modification.
> Approved by:              If you needed approval for this commit.
> Obtained from:            If the change is from a third party.
> MFC after:                N [day[s]|week[s]|month[s]].  Request a reminder email.
> MFH:                      Ports tree branch name.  Request approval for merge.
> Relnotes:                 Set to 'yes' for mention in release notes.
> Security:                 Vulnerability reference (one per line) or description.
> Sponsored by:             If the change was sponsored by an organization (each collaborator).
> Differential Revision:    https://reviews.freebsd.org/D### (*full* phabric URL needed).
> Empty fields above will be automatically removed.

M    sys/powerpc/aim/aim_machdep.c
M    sys/powerpc/aim/trap_subr32.S
M    sys/powerpc/aim/trap_subr64.S
M    sys/powerpc/include/trap.h

4 years agoGEOM label: strip leading/trailing space synthesizing devfs names
cem [Sat, 18 Jan 2020 03:33:44 +0000 (03:33 +0000)]
GEOM label: strip leading/trailing space synthesizing devfs names

%20%20%20 is ugly and doesn't really help make human-readable devfs names.

PR: 243318
Reported by: Peter Eriksson <pen AT lysator.liu.se>
Relnotes: yes

4 years agoAdd a 'SINGLETON' directive to kobj interface definition
jhibbits [Sat, 18 Jan 2020 02:39:38 +0000 (02:39 +0000)]
Add a 'SINGLETON' directive to kobj interface definition

Summary:
This makes the interface described in the definition file act like a
pseudo-IFUNC service, by caching the found method locally.

Applying this to the PowerPC MMU definitions, it yields a significant
(15-20%) performance improvement, seen in both a 'make buildworld' and a
parallel build of LLVM, on a POWER9 system.

Reviewed By: imp
Differential Revision: https://reviews.freebsd.org/D23245

4 years agovfs: distribute freevnodes counter per-cpu
mjg [Sat, 18 Jan 2020 01:29:02 +0000 (01:29 +0000)]
vfs: distribute freevnodes counter per-cpu

It gets rolled up to the global when deferred requeueing is performed.
A dedicated read routine makes sure to return a value only off by a certain
amount.

This soothes a global serialisation point for all 0<->1 hold count transitions.

Reviewed by: jeff
Differential Revision: https://reviews.freebsd.org/D23235

4 years agopowerpc: Fix the NUMA domain list on powernv
jhibbits [Sat, 18 Jan 2020 01:26:54 +0000 (01:26 +0000)]
powerpc: Fix the NUMA domain list on powernv

Summary:
Consolidate the NUMA associativity handling into a platform function.
Non-NUMA platforms will just fall back to the default (0).  Currently
only implemented for powernv, which uses a lookup table to map the
device tree associativity into a system NUMA domain.

Fixes hangs on powernv after r356534, and corrects a fairly longstanding
bug in powernv's NUMA handling, which ended up using domains 1 and 2 for
devices and memory on power9, while CPUs were bound to domains 0 and 1.

Reviewed by: bdragon, luporl
Differential Revision: https://reviews.freebsd.org/D23220

4 years ago[PowerPC] Fix Book-E direct map for >=16G ram on e5500
bdragon [Sat, 18 Jan 2020 01:22:54 +0000 (01:22 +0000)]
[PowerPC] Fix Book-E direct map for >=16G ram on e5500

It turns out the maximum TLB1 page size on e5500 is 4G, despite the format
being defined for up to 1TB.

So, we need to clamp the DMAP TLB1 entries to not attempt to create 16G or
larger entries.

Fixes boot on my X5000 in which I just installed 16G of RAM.

Reviewed by: jhibbits
Sponsored by: Tag1 Consulting, Inc.
Differential Revision: https://reviews.freebsd.org/D23244

4 years ago[PowerPC] Save a dword in the powerpc64 signal trampoline
bdragon [Fri, 17 Jan 2020 23:41:35 +0000 (23:41 +0000)]
[PowerPC] Save a dword in the powerpc64 signal trampoline

In r291668, an instruction was added to sigcode64.S without the nop pad at
the end being taken out.

Due to alignment, this means that a dword is being wasted on the shared
page for no reason.

Take out this nop, and add some comments while I'm here.

Reviewed by: jhibbits
Sponsored by: Tag1 Consulting, Inc.
Differential Revision: https://reviews.freebsd.org/D23055

4 years agoFix regression in bsdinstall post r356740 - partedit errno(2) 21 EISDIR
woodsb02 [Fri, 17 Jan 2020 22:26:41 +0000 (22:26 +0000)]
Fix regression in bsdinstall post r356740 - partedit errno(2) 21 EISDIR

This resulted in the partitioning step failing if either of the
"Auto (UFS)" or "Manual" options were selected.

Reason: partedit was attempting to open a directory (TMPDIR) read/write,
which resulted in errno(2) 21 - EISDIR - Is a directory.

Reported by: Clay Daniels <clay.daniels.jr@gmail.com>
Reviewed by: Ryan Moeller <ryan@freqlabs.com>
Approved by: emaste, bcran
Differential Revision: https://reviews.freebsd.org/D23232

4 years agonet80211: Move rate printing in amrr_node_stats() to a separate method
cem [Fri, 17 Jan 2020 22:04:11 +0000 (22:04 +0000)]
net80211: Move rate printing in amrr_node_stats() to a separate method

This makes amrr_node_stats() cleaner and allows the rate printing to be
reusable.

Submitted by: Neel Chauhan <neel at neelc.org>
Reviewed by: adrian
Differential Revision: https://reviews.freebsd.org/D22318

4 years agobcm2835_vcbus: unifdef all platform definitions
kevans [Fri, 17 Jan 2020 21:39:28 +0000 (21:39 +0000)]
bcm2835_vcbus: unifdef all platform definitions

Raspberry Pi are all over the board, and the reality is that there's no harm
in including all of the definitions by default but plenty of harm in the
current situation. This change is safe because we match a definition by root
/compatible in the FDT, so there will be no false-positives because of it.

The main array of definitions grows, but it's only walked exactly once to
determine which we need to use.

4 years agoCheck for invalid sstatus values in set_mcontext().
jhb [Fri, 17 Jan 2020 19:13:49 +0000 (19:13 +0000)]
Check for invalid sstatus values in set_mcontext().

Previously, this check was only in sys_sigreturn() which meant that
user applications could write invalid values to the register via
setcontext() or swapcontext().

Reviewed by: mhorne
Sponsored by: DARPA
Differential Revision: https://reviews.freebsd.org/D23219

4 years agoSave and restore floating point registers in get/set_mcontext().
jhb [Fri, 17 Jan 2020 19:01:59 +0000 (19:01 +0000)]
Save and restore floating point registers in get/set_mcontext().

arm64 and riscv were only saving and restoring floating point
registers for sendsig() and sys_sigreturn(), but not for getcontext(),
setcontext(), and swapcontext().

While here, remove an always-false check for uap being NULL from
sys_sigreturn().

Reviewed by: br, mhorne
Sponsored by: DARPA
Differential Revision: https://reviews.freebsd.org/D23218

4 years agosrc.conf.5: regen after r356836, DMAGENT dependency on OPENSSL
emaste [Fri, 17 Jan 2020 17:56:31 +0000 (17:56 +0000)]
src.conf.5: regen after r356836, DMAGENT dependency on OPENSSL

4 years agosrc.opts.mk: force DMAGENT off under WITHOUT_OPENSSL
emaste [Fri, 17 Jan 2020 17:53:13 +0000 (17:53 +0000)]
src.opts.mk: force DMAGENT off under WITHOUT_OPENSSL

dma(8) depends on OpenSSL unconditionally.

Reported by: Michael Dexter's Build Options Survey run
MFC after: 1 weeks
Sponsored by: The FreeBSD Foundation

4 years agoRISC-V: fix global pointer assignment at boot
mhorne [Fri, 17 Jan 2020 17:03:25 +0000 (17:03 +0000)]
RISC-V: fix global pointer assignment at boot

As part of the RISC-V ABI, the gp register is expected to initialized
with the address of __global_pointer$ as early as possible. This allows
loads and stores from .sdata to be relaxed based on the value of gp. In
locore.S we do this initialization twice, once each for va and mpva.
However, in both cases the initialization is preceded by an la
instruction, which in theory could be relaxed by the linker.

Move the initialization of gp to be slightly earlier (before la
cpu_exception_handler), and add an additional gp initialization at the
very beginning of _start, before virtual memory is set up.

Reported by: jrtc27
Reviewed by: jrtc27
Differential Revision: https://reviews.freebsd.org/D23139

4 years agoUse unsigned loads in fubyte, fuword16, generic_bs_r_1, generic_bs_r_2
br [Fri, 17 Jan 2020 16:48:20 +0000 (16:48 +0000)]
Use unsigned loads in fubyte, fuword16, generic_bs_r_1, generic_bs_r_2
as these functions should do zero-extend.

Discovered by running pci_read_cap(), and by hint from James Clarke.

Reviewed by: James Clarke <jrtc27@jrtc27.com>
Sponsored by: DARPA, AFRL
Differential Revision: https://reviews.freebsd.org/D23236

4 years agolibc: fix build after r356830
mjg [Fri, 17 Jan 2020 15:45:39 +0000 (15:45 +0000)]
libc: fix build after r356830

Apparently building with 'cd lib/libc; make all install' is not the same
as buildworld.

Reported by: Michael Butler

4 years ago[PPC] Fix wrong comment
luporl [Fri, 17 Jan 2020 14:43:58 +0000 (14:43 +0000)]
[PPC] Fix wrong comment

pcb_context[20] holds r12-r31 and not r14-r31, as the comment said.

4 years agovfs: provide F_ISUNIONSTACK as a kludge for libc
mjg [Fri, 17 Jan 2020 14:42:25 +0000 (14:42 +0000)]
vfs: provide F_ISUNIONSTACK as a kludge for libc

Prior to introduction of this op libc's readdir would call fstatfs(2), in
effect unnecessarily copying kilobytes of data just to check fs name and a
mount flag.

Reviewed by: kib (previous version)
Differential Revision: https://reviews.freebsd.org/D23162

4 years agolibc: assume no union stack if fstatfs fails in readdir
mjg [Fri, 17 Jan 2020 14:40:09 +0000 (14:40 +0000)]
libc: assume no union stack if fstatfs fails in readdir

The failure is not really expected, but should it happen it's better to
get some data.

Suggested by: kib

4 years agovfs: shorten lock hold time in vdbatch_process
mjg [Fri, 17 Jan 2020 14:39:00 +0000 (14:39 +0000)]
vfs: shorten lock hold time in vdbatch_process

4 years agoMETA_MODE: Allow 'make all install' to work with filemon.
bdrewery [Fri, 17 Jan 2020 14:29:27 +0000 (14:29 +0000)]
META_MODE: Allow 'make all install' to work with filemon.

Filemon will add the ability to ignore the cookie if the installed file is
missing. Without filemon that's not possible though so if the cookie is present
an the command unchanged then the install wouldn't run.

Sponsored by: DellEMC
MFC after: 2 weeks

4 years agoChange argument order of epoch_call() to more natural, first function,
glebius [Fri, 17 Jan 2020 06:10:24 +0000 (06:10 +0000)]
Change argument order of epoch_call() to more natural, first function,
then its argument.

Reviewed by: imp, cem, jhb

4 years agoFix a long standing bug that was made worse in r355765. When we are cowing a
jeff [Fri, 17 Jan 2020 03:44:04 +0000 (03:44 +0000)]
Fix a long standing bug that was made worse in r355765.  When we are cowing a
page that was previously mapped read-only it exists in pmap until pmap_enter()
returns.  However, we held no reference to the original page after the copy
was complete.  This allowed vm_object_scan_all_shadowed() to collapse an
object that still had pages mapped.  To resolve this, add another page pointer
to the faultstate so we can keep the page xbusy until we're done with
pmap_enter().  Handle busy pages in scan_all_shadowed.  This is already done
in vm_object_collapse_scan().

Reviewed by: kib, markj
Differential Revision: https://reviews.freebsd.org/D23155

4 years agoSmall tweak to the default behavior of shutdown -c
imp [Fri, 17 Jan 2020 01:20:48 +0000 (01:20 +0000)]
Small tweak to the default behavior of shutdown -c

'shutdown -c' is supposed to power cycle the system rather than doing a normal
reboot. However, when that fails, it halts the system. This is not quite right
since the intent isn't to halt the system but to restart. Make the default init
behavior be to restart the system. The halt(8) interface can be used if you'd
like to powercycle or halt.

MFC After: 1 week
Differential Revision: https://reviews.freebsd.org/D23129

4 years agoWe only want to send the speedup to the lower layers when there's a shortage.
imp [Fri, 17 Jan 2020 01:16:23 +0000 (01:16 +0000)]
We only want to send the speedup to the lower layers when there's a shortage.

Only send a speedup when there's a shortage. While this is a little racy, lost
races aren't a big deal for this function. If there's a shorage just popping up
after we check these values, then we'll catch it next time. If there's a
shortage that's just clearing up, we may do some work at the lower layers a
little sooner than we otherwise would have. Sicne shortages are relatively rare
events, both races are acceptable.

Reviewed by: chs
Differential Revision: https://reviews.freebsd.org/D23182

4 years agoUse buf to send speedup
imp [Fri, 17 Jan 2020 01:16:19 +0000 (01:16 +0000)]
Use buf to send speedup

It turns out there's a problem with using g_io to send the speedup. It leads to
a race when there's a resource shortage when a disk fails.

Instead, send BIO_SPEEDUP via struct buf. This is pretty straight forward,
except we need to transfer the bio_flags from b_ioflags for BIO_SPEEDUP commands
in g_vfs_strategy.

Reviewed by: kirk, chs
Differential Revision: https://reviews.freebsd.org/D23117

4 years agoPass BIO_SPEEDUP through all the geom layers
imp [Fri, 17 Jan 2020 01:15:55 +0000 (01:15 +0000)]
Pass BIO_SPEEDUP through all the geom layers

While some geom layers pass unknown commands down, not all do. For the ones that
don't, pass BIO_SPEEDUP down to the providers that constittue the geom, as
applicable. No changes to vinum or virstor because I was unsure how to add this
support, and I'm also unsure how to test these. gvinum doesn't implement
BIO_FLUSH either, so it may just be poorly maintained. gvirstor is for testing
and not supportig BIO_SPEEDUP is fine.

Reviewed by: chs
Differential Revision: https://reviews.freebsd.org/D23183

4 years agounionfs: use MNTK_NOMSYNC
mjg [Thu, 16 Jan 2020 22:45:08 +0000 (22:45 +0000)]
unionfs: use MNTK_NOMSYNC

4 years agoFix pfdenied not returning any results
kp [Thu, 16 Jan 2020 22:08:05 +0000 (22:08 +0000)]
Fix pfdenied not returning any results

When _a is empty we end up with an invalid invocation of pfctl, and no output.
We must add quotes to make it clear to pfctl that we're passing an empty anchor
name.

PR: 224415
Submitted by: sigsys AT gmail.com
MFC after: 2 weeks

4 years agodwmmc: Remove max_hz from the softc
manu [Thu, 16 Jan 2020 21:50:53 +0000 (21:50 +0000)]
dwmmc: Remove max_hz from the softc

We never use it so directly set the value to the mmc host structure.

4 years agovfs: increment numvnodes without the vnode list lock unless under pressure
mjg [Thu, 16 Jan 2020 21:45:21 +0000 (21:45 +0000)]
vfs: increment numvnodes without the vnode list lock unless under pressure

The vnode list lock is only needed to reclaim free vnodes or kick the vnlru
thread (or to block and not miss a wake up (but note the sleep has a timeout so
this would not be a correctness issue)). Try to get away without the lock by
just doing an atomic increment.

The lock is contended e.g., during poudriere -j 104 where about half of all
acquires come from vnode allocation code.

Note the entire scheme needs a rewrite, the above just reduces it's SMP impact.

Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D23140

4 years agovfs: refcator vnode allocation
mjg [Thu, 16 Jan 2020 21:43:13 +0000 (21:43 +0000)]
vfs: refcator vnode allocation

Semantics are almost identical. Some code is deduplicated and there are
fewer memory accesses.

Reviewed by: kib, jeff
Differential Revision: https://reviews.freebsd.org/D23158

4 years agorandom(6): Fix off-by-one
cem [Thu, 16 Jan 2020 21:38:44 +0000 (21:38 +0000)]
random(6): Fix off-by-one

After r355693, random(6) -f sometimes fail to output all the lines of the
input file.  This is because the range from which random indices are chosen
is too big, so occasionally the random selection doesn't correspond to any
line and nothing gets printed.

(Ed. note: Mea culpa.  Working on r355693, I was confused by the sometime
use of 1-indexing, sometimes 0-indexing in randomize_fd().)

Submitted by: Ryan Moeller <ryan AT freqlabs.com>
X-MFC-With: r355693
Sponsored by: iXsystems, Inc.
Differential Revision: https://reviews.freebsd.org/D23199