]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
4 years agoRemove sparc64 kernel support
imp [Mon, 3 Feb 2020 17:35:11 +0000 (17:35 +0000)]
Remove sparc64 kernel support

Remove all sparc64 specific files
Remove all sparc64 ifdefs
Removee indireeect sparc64 ifdefs

4 years agoRemove sparc64 support from the boot loader.
imp [Mon, 3 Feb 2020 17:34:57 +0000 (17:34 +0000)]
Remove sparc64 support from the boot loader.

Remove all the sparc64 specific bits, both files and ifdefs.

4 years agoUnblock kstat.zfs.misc.dbufstats sysctls.
mav [Mon, 3 Feb 2020 17:10:40 +0000 (17:10 +0000)]
Unblock kstat.zfs.misc.dbufstats sysctls.

It is not so much broken to hide it after we wasted time to collect it.

MFC after: 2 weeks
Sponsored by: iXsystems, Inc.

4 years agocapsicum: faster cap_rights_contains
mjg [Mon, 3 Feb 2020 17:08:11 +0000 (17:08 +0000)]
capsicum: faster cap_rights_contains

Instead of doing a 2 iteration loop (determined at runeimt), take advantage
of the fact that the size is already known.

While here provdie cap_check_inline so that fget_unlocked does not have to
do a function call.

Verified with the capsicum suite /usr/tests.

4 years agotmpfs: inline tmpfs_update
mjg [Mon, 3 Feb 2020 17:06:21 +0000 (17:06 +0000)]
tmpfs: inline tmpfs_update

It was generated to be just a jumping off point to tmpfs_itimes.

While here provide a dedicated variant for getattr since we normally don't
expect to need to the update from that caller.

4 years agoaddr2line: Cache CU DIEs upon a successful address lookup.
markj [Mon, 3 Feb 2020 16:41:40 +0000 (16:41 +0000)]
addr2line: Cache CU DIEs upon a successful address lookup.

Previously, addr2line would sequentially search all CUs for each input
address.  For some uses, notably syzkaller's code coverage map generator,
this was extremely slow.  Add a CU cache into which entries are added
following a successful lookup, and search the cache before falling back
to a scan.  When translating a large number of addresses this yields
slightly better performance than GNU addr2line.

Garbage-collect an unused hash table which appears to have been intended
for the same purpose.  A hash table doesn't seem particularly suitable
since each CU spans a range of addresses.

Submitted by: Tiger Gao <tig@freebsdfoundation.org>
MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D23418

4 years agounits(1): Refactor the manual page and update usage information
0mp [Mon, 3 Feb 2020 15:22:46 +0000 (15:22 +0000)]
units(1): Refactor the manual page and update usage information

Changes to units.1:

- Change the description to a more descriptive "conversion calculator".
- Sort options.
- Split the description into sections to make it easier to navigate the
  manual page.
- Improve the description of various options.
- Document the default value of the output format.
- Use more mdoc macros for better readability.
- Document the behavior of the PATH environmental variable.
- Improve examples.
- Add sections: EXIT STATUS, DIAGNOSTICS, and HISTORY.
- Document that units(1) cannot convert negative values and it handles long
  unit lists poorly.
- Update the documentation of the -V flag to match the implementation.
  units(1) prints its version and the units data file instead of its
  version and usage information.

Changes to units.c:

- Update usage information.
- Sort longopts elements.

This commit does not attempts to change the current behavior of units(1).
What's left to do is probably defining a better versioning (at the moment
units(1) always reports "FreeBSD units" as its version) and changing the
behavior of the -V flag to only print version.

Reviewed by: allanjude (earlier version), bcr
Approved by: bcr (mentor)
Differential Revision: https://reviews.freebsd.org/D18977

4 years agoRemove the GICv3 ITS irq and replace it with an ID
andrew [Mon, 3 Feb 2020 14:38:19 +0000 (14:38 +0000)]
Remove the GICv3 ITS irq and replace it with an ID

In r357324 most of the use of gi_irq was moved to gi_lpi. Complete this
with the last few places we need the IRQ value and create gi_id for the
per-device value we need.

MFC after: 2 weeks
Sponsored by: DARPA, AFRL

4 years agofd: fix f_count acquire in fget_unlocked
mjg [Mon, 3 Feb 2020 14:28:31 +0000 (14:28 +0000)]
fd: fix f_count acquire in fget_unlocked

The code was using a hand-rolled fcmpset loop, while in other places the same
count is manipulated with the refcount API.

This transferred from a stylistic issue into a bug after the API got extended
to support flags. As a result the hand-rolled loop could bump the count high
enough to set the bit flag. Another bump + refcount_release would then free
the file prematurely.

The bug is only present in -CURRENT.

4 years agoFix up various vnode-related asserts which did not dump the used vnode
mjg [Mon, 3 Feb 2020 14:25:32 +0000 (14:25 +0000)]
Fix up various vnode-related asserts which did not dump the used vnode

4 years agoUse a unique name for the GICv3 ITS vmem
andrew [Mon, 3 Feb 2020 13:50:55 +0000 (13:50 +0000)]
Use a unique name for the GICv3 ITS vmem

When there are multiple GICv3 ITS devices we don't know which vmem is for
which device. Use device_get_nameunit to get a per-device name.

MFC after: 2 weeks
Sponsored by: DARPA, AFRL

4 years agoremove stale gnu/lib/csu directory
emaste [Mon, 3 Feb 2020 13:50:15 +0000 (13:50 +0000)]
remove stale gnu/lib/csu directory

As of r357338 gnu/lib/csu is never used.

4 years agoDisable the use of the quantum cache in the GICv3 ITS
andrew [Mon, 3 Feb 2020 13:47:41 +0000 (13:47 +0000)]
Disable the use of the quantum cache in the GICv3 ITS

This uses UMA to allocate space. It causes issues when there are multiple
ITS devices in the system where interrupts are not allocated from a low
address on some interrupt controllers. Disabling the quantum cache fixes
this on the Neoverse N1 SDP.

MFC after: 2 weeks
Sponsored by: DARPA, AFRL

4 years agoloader: bc_add can not use any other probes than ah=0x4b
tsoome [Mon, 3 Feb 2020 11:33:33 +0000 (11:33 +0000)]
loader: bc_add can not use any other probes than ah=0x4b

CD boot is broken for some systems since bioscd and biosdisk merge. The issue is that we can not use anything else than int 13 ah=0x4b to query cd information.

The patch does restore the same probe as was originally used in bioscd.c. Additionally extra buffer padding is used to avoid memory corruption caused by some systems.

PR: 234031
Reported by: ultramage and others
MFC after: 1 day

4 years agoOn powerpc, we use ofw_syscons for device sc. That references the default
imp [Mon, 3 Feb 2020 05:38:45 +0000 (05:38 +0000)]
On powerpc, we use ofw_syscons for device sc. That references the default
fonts. As a workaround, remove the static. vt is default on powerpc, but there's
a few old macs that still fail with vt. sc is used as a work arouond for those
machines, and the kernel fails to build w/o it.

4 years agonetinet/libalias: Fix typo in debug message
cem [Mon, 3 Feb 2020 05:19:44 +0000 (05:19 +0000)]
netinet/libalias: Fix typo in debug message

No functional change.

PR: 243831
Submitted by: Neel Chauhan <neel AT neelc DOT org>
Differential Revision: https://reviews.freebsd.org/D23365

4 years agotypo: Registration.
pfg [Mon, 3 Feb 2020 02:02:13 +0000 (02:02 +0000)]
typo: Registration.

Pointed by: Dikshie Fauzie

4 years agoethernet: Minor cleanup.
pfg [Mon, 3 Feb 2020 01:08:15 +0000 (01:08 +0000)]
ethernet: Minor cleanup.
Consistently use uppercase for ethertype hex numbers.

4 years agoacpi_ibm: remove superfluous cast
emaste [Sun, 2 Feb 2020 20:56:18 +0000 (20:56 +0000)]
acpi_ibm: remove superfluous cast

Reported by: kib

4 years agoDiff reduction against NetBSD, no functional change.
delphij [Sun, 2 Feb 2020 20:53:31 +0000 (20:53 +0000)]
Diff reduction against NetBSD, no functional change.

MFC after: 1 week

4 years agoRemove unused include.
delphij [Sun, 2 Feb 2020 20:51:28 +0000 (20:51 +0000)]
Remove unused include.

MFC after: 2 weeks

4 years agolibc: provide fputc_unlocked
kevans [Sun, 2 Feb 2020 19:45:12 +0000 (19:45 +0000)]
libc: provide fputc_unlocked

Among the same justification as the other stdio _unlocked; in addition to an
inline version in <stdio.h>, we must provide a function in libc as well for
the functionality. This fixes the lang/gcc* builds, which want to use the
symbol from libc.

PR: 243810
Reported by: antoine, swills, Michael <michael.adm gmail com>
X-MFC-With: r357284

4 years ago[PowerPC] Fix feholdexcept() on powerpc referencing the wrong struct member
bdragon [Sun, 2 Feb 2020 19:16:52 +0000 (19:16 +0000)]
[PowerPC] Fix feholdexcept() on powerpc referencing the wrong struct member

In r355656, endianness handling of the floating point environment was fixed
in the PowerPC code to work as intended.

However, one bit got missed, causing feholdexcept() to mis-save the fenv.

Submitted by: Renato Riolino <renato.riolino@eldorado.org.br>
Differential Revision: https://reviews.freebsd.org/D23382

4 years agostyle(9): Fix spaces after #define.
pfg [Sun, 2 Feb 2020 19:02:07 +0000 (19:02 +0000)]
style(9): Fix spaces after #define.

No functional change.

4 years agoacpi_ibm: whitespace and wrapping cleanup
emaste [Sun, 2 Feb 2020 19:01:16 +0000 (19:01 +0000)]
acpi_ibm: whitespace and wrapping cleanup

4 years agoethernet: add some more Ethertypes.
pfg [Sun, 2 Feb 2020 18:33:20 +0000 (18:33 +0000)]
ethernet: add some more Ethertypes.

Sort ETHERTYPE_FCOE, from r357414.

4 years agoethernet: add some more Ethertypes.
pfg [Sun, 2 Feb 2020 18:27:37 +0000 (18:27 +0000)]
ethernet: add some more Ethertypes.

Add some types based on other BSDs and also add EtherCat and PROFINET, which
are IEC standards.

There is a public list (CSV format) at:
https://standards.ieee.org/products-services/regauth/

MFC after: 2 weeks

4 years agoacpi_ibm: whitespace fixup
emaste [Sun, 2 Feb 2020 18:07:47 +0000 (18:07 +0000)]
acpi_ibm: whitespace fixup

4 years agoProvide O_SEARCH
kevans [Sun, 2 Feb 2020 16:34:57 +0000 (16:34 +0000)]
Provide O_SEARCH

O_SEARCH is defined by POSIX [0] to open a directory for searching, skipping
permissions checks on the directory itself after the initial open(). This is
close to the semantics we've historically applied for O_EXEC on a directory,
which is UB according to POSIX. Conveniently, O_SEARCH on a file is also
explicitly undefined behavior according to POSIX, so O_EXEC would be a fine
choice. The spec goes on to state that O_SEARCH and O_EXEC need not be
distinct values, but they're not defined to be the same value.

This was pointed out as an incompatibility with other systems that had made
its way into libarchive, which had assumed that O_EXEC was an alias for
O_SEARCH.

This defines compatibility O_SEARCH/FSEARCH (equivalent to O_EXEC and FEXEC
respectively) and expands our UB for O_EXEC on a directory. O_EXEC on a
directory is checked in vn_open_vnode already, so for completeness we add a
NOEXECCHECK when O_SEARCH has been specified on the top-level fd and do not
re-check that when descending in namei.

[0] https://pubs.opengroup.org/onlinepubs/9699919799/

Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D23247

4 years agozfs: light refactor to indicate cachedlookup in zfs_lookup
kevans [Sun, 2 Feb 2020 16:10:33 +0000 (16:10 +0000)]
zfs: light refactor to indicate cachedlookup in zfs_lookup

If we come from VOP_CACHEDLOOKUP, we must skip the VEXEC check as it will
have been done in the caller (vfs_cache_lookup). This is a part of D23247,
which may skip the earlier VEXEC check as well if the root fd was opened
with O_SEARCH.

This one required slightly more work as zfs_lookup may also be called
indirectly as VOP_LOOKUP or a couple of other places where we must do the
check.

4 years agopseudofs: don't do VEXEC check in VOP_CACHEDLOOKUP
kevans [Sun, 2 Feb 2020 15:36:12 +0000 (15:36 +0000)]
pseudofs: don't do VEXEC check in VOP_CACHEDLOOKUP

VOP_CACHEDLOOKUP should assume that the appropriate VEXEC check has been
done in the caller (vfs_cache_lookup), so it does not belong here.

4 years agoMove ce enable to SOURCELESS_HOST
emaste [Sun, 2 Feb 2020 14:41:09 +0000 (14:41 +0000)]
Move ce enable to SOURCELESS_HOST

ce contains obfuscated code that runs on the host's processor

4 years agoPer the firm plan, start to remove sparc64
imp [Sun, 2 Feb 2020 11:37:27 +0000 (11:37 +0000)]
Per the firm plan, start to remove sparc64

The sparc64 architecture is being removed from FreeBSD 13, starting
now. This removes it from the top level only. It is the only
architecture that didn't see substantial work after the call to get
things working with the external toolchain.

4 years agofd: sprinkle some predits around fget
mjg [Sun, 2 Feb 2020 09:38:40 +0000 (09:38 +0000)]
fd: sprinkle some predits around fget

clang inlines fget -> _fget into kern_fstat and eliminates several checkes,
but prior to this change it would assume fget_unlocked was likely to fail
and consequently avoidable jumps got generated.

4 years agofd: use atomic_load_ptr instead of hand-rolled cast through volatile
mjg [Sun, 2 Feb 2020 09:37:16 +0000 (09:37 +0000)]
fd: use atomic_load_ptr instead of hand-rolled cast through volatile

No change in assembly.

4 years agovfs: remove the now empty vop_unlock_post
mjg [Sun, 2 Feb 2020 09:36:32 +0000 (09:36 +0000)]
vfs: remove the now empty vop_unlock_post

4 years agovfs: remove the never set VDESC_VPP_WILLRELE flag
mjg [Sun, 2 Feb 2020 09:35:48 +0000 (09:35 +0000)]
vfs: remove the never set VDESC_VPP_WILLRELE flag

4 years agoMove font.h generation to conf/files from conf/files.*
imp [Sun, 2 Feb 2020 08:27:26 +0000 (08:27 +0000)]
Move font.h generation to conf/files from conf/files.*

Use ${SRCTOP} instead of /usr/share.
Prefer to depend on option sc_dflt_fnt instead of sc.
gc the 4 otherwise identical instances in the tree.
Platforms that don't need this won't included it.

4 years agoFix old-style build
imp [Sun, 2 Feb 2020 08:27:20 +0000 (08:27 +0000)]
Fix old-style build

Fix the old-style build by using ${SRCTOP} instead of a weird
construct that only works for new-style build.
Simplify the building of keymap files by using macros
Move atkbdmap.h in files.x86
This has been broken since r296899 which removed the implicit
dependency on /usr/share.

4 years agoKill old armv4 busdma
imp [Sun, 2 Feb 2020 08:27:14 +0000 (08:27 +0000)]
Kill old armv4 busdma

Move to having one busdma option for arm: the armv6/v7 one.
Kill now-unused option ARM_USE_V6_BUSDMA too.
Fixup files.arm to match rename.

4 years agoMove arm back to having one LINT
imp [Sun, 2 Feb 2020 08:27:08 +0000 (08:27 +0000)]
Move arm back to having one LINT

Now that armv5 is gone, we no longer need multiple LINT files. Kill
the odd-ball support here. From now on, we just have LINT built from
notes like all the other platforms. Keep the removal of LINT-V5/7
to remove stale files for a while still..

4 years agoRemove old boardid/mach-types support.
imp [Sun, 2 Feb 2020 06:52:10 +0000 (06:52 +0000)]
Remove old boardid/mach-types support.

This has been long obsolete in linux and now that all armv4/5 support
is gone, it can be retired too.

4 years agoRemove armv5 marvell support.
imp [Sun, 2 Feb 2020 06:52:04 +0000 (06:52 +0000)]
Remove armv5 marvell support.

Per plans to remove armv5 support, remove the armv5 marvell discovery,
kirkwood and orion families.

4 years agoRemove Ralink RT1310 support
imp [Sun, 2 Feb 2020 06:51:42 +0000 (06:51 +0000)]
Remove Ralink RT1310 support

This armv5 SoC is being removed consistent with the removal of
armv5 support, per discussions in arm@ mailing list.

4 years agoRemove vpo.4
imp [Sun, 2 Feb 2020 04:53:27 +0000 (04:53 +0000)]
Remove vpo.4

The Parallel Port SCSI adapter was interesting for 100MB ZIP drives, but is no
longer used or maintained. Remove it from the tree.

The Parallel Port microsequencer (microseq.9) is now mostly unused in the tree,
but remains. PPI still refrences it, but doesn't use its full functionality.

Relnotes: Yes
Reviewed by: rgrimes@, Ihor Antonov
Discussed on: arch@
Differential Revision:  https://reviews.freebsd.org/D23389

4 years agoAdd deprecation notice to vpo.4
imp [Sun, 2 Feb 2020 04:52:28 +0000 (04:52 +0000)]
Add deprecation notice to vpo.4

This driver has seen no real changes for almost 20 years. It's for
hardware that's 25 years old. It has no reports of active use, nor
has it been seen in the NYCBug dmesg database at all. Schedule
its removal for 13.0.

Reviewed by: rgrimes@ (earlier version)
Relnote: Yes
MFC After: 3 days
Differential Revision: https://reviews.freebsd.org/D23403

4 years agoFix a bug in r356776 where the page allocator was not properly restored to
jeff [Sat, 1 Feb 2020 23:46:30 +0000 (23:46 +0000)]
Fix a bug in r356776 where the page allocator was not properly restored to
the percpu page allocator after it had been temporarily overridden by
startup_alloc.

Reported by: pho, bdragon

4 years agoFix build.
kib [Sat, 1 Feb 2020 23:16:30 +0000 (23:16 +0000)]
Fix build.

Sponsored by: The FreeBSD Foundation

4 years agoIn dmar_gas_lowermatch, skip searching a subtree if all its addresses are greater...
dougm [Sat, 1 Feb 2020 21:47:34 +0000 (21:47 +0000)]
In dmar_gas_lowermatch, skip searching a subtree if all its addresses are greater than lowaddr.
In dmar_gas_uppermatch, skip searching a subtree if all its gaps-between-alloctions are too small.

Reviewed by: kib
Tested by: pho
Differential Revision: https://reviews.freebsd.org/D23391

4 years agoAmend r357367 by using register values from the TPM datasheet.
dim [Sat, 1 Feb 2020 21:43:45 +0000 (21:43 +0000)]
Amend r357367 by using register values from the TPM datasheet.

As Ian Lepore noted, writing ~1 to a register might have a completely
different effect than doing a regular read-modify-write operation.

Follow the TCG_PC_Client_Platform_TPM_Profile_PTP_2.0_r1.03_v22
datasheet instead, and use the actual values mentioned there:
(uint32_t)1 to cancel the command, (uint32_t)0 to clear the field.

MFC after: 3 days

4 years agocache: replace kern___getcwd with vn_getcwd
mjg [Sat, 1 Feb 2020 20:38:38 +0000 (20:38 +0000)]
cache: replace kern___getcwd with vn_getcwd

The previous routine was resulting in extra data copies most notably in
linux_getcwd.

4 years agotpm(4): Fix 'go ready' in TPM 2.0 TIS driver
cem [Sat, 1 Feb 2020 20:38:22 +0000 (20:38 +0000)]
tpm(4): Fix 'go ready' in TPM 2.0 TIS driver

tpmtis_go_ready() read the value of the TPM_STS register, ORed
TPM_STS_CMD_READY with it, and wrote it back.  However, the TPM Profile
(PTP) specification states that only one bit in the write request value may
be set to 1, or else the entire write request is ignored.

Fix by just writing TPM_STS_CMD_READY.

Similarly, remove the call which clears the TPM_STS_CMD_READY flag in the
same function.  It was being ignored for the same reason.

Submitted by: Darrick Lew <darrick.freebsd AT gmail.com>
Reviewed by: vangyzen, myself
MFC after: if you care about stable, you might want to do so
Sponsored by: Dell EMC Isilon
Differential Revision: https://reviews.freebsd.org/D23081

4 years agocache: return the total length from vn_fullpath1
mjg [Sat, 1 Feb 2020 20:37:11 +0000 (20:37 +0000)]
cache: return the total length from vn_fullpath1

This removes strlen from getcwd.

4 years agocache: remove vnode -> path lookup disablement
mjg [Sat, 1 Feb 2020 20:36:35 +0000 (20:36 +0000)]
cache: remove vnode -> path lookup disablement

It seems to be of little to no use even when debugging.

Interested parties can resurrect it and gate compilation with a macro.

4 years agovfs: consistently use size_t for buflen around VOP_VPTOCNP
mjg [Sat, 1 Feb 2020 20:34:43 +0000 (20:34 +0000)]
vfs: consistently use size_t for buflen around VOP_VPTOCNP

4 years agorand(3): Replace implementation with one backed by random(3) algorithm
cem [Sat, 1 Feb 2020 20:33:23 +0000 (20:33 +0000)]
rand(3): Replace implementation with one backed by random(3) algorithm

rand(3)'s standard C API is extremely limiting, but we can do better
than the historical 32-bit state Park-Miller LCG we've shipped since
2001: r73156.

The justification provided at the time for not using random(3) was that
rand_r(3) could not be made to use the same algorithm.  That is still
true.  However, the irrelevance of rand_r(3) is increasingly obvious.
Since that time, POSIX has marked the interface obsolescent.  rand_r(3)
never became part of the standard C library.  If not for API
compatibility reasons, I would just remove rand_r(3) entirely.

So, I do not believe it is a problem for rand_r(3) and rand(3) to
diverge.

The 12 ABI is maintained with compatibility definitions, but this
revision does subtly change the API of rand(3).  The sequences of
pseudorandom numbers produced in programs built against new versions of
libc will differ from programs built against prior versions of libc.

Reviewed by: kevans, markm
MFC after: no
Relnotes: yes
Differential Revision: https://reviews.freebsd.org/D23290

4 years agohwpstate_intel(4): Save admin-set EPP/EPB and restore after suspend
cem [Sat, 1 Feb 2020 20:12:02 +0000 (20:12 +0000)]
hwpstate_intel(4): Save admin-set EPP/EPB and restore after suspend

4 years agohwpstate_intel(4): Print failure message only on failure
cem [Sat, 1 Feb 2020 20:11:25 +0000 (20:11 +0000)]
hwpstate_intel(4): Print failure message only on failure

X-MFC-With: r357379

4 years agohwpstate_intel(4): Detect and support PKG variant
cem [Sat, 1 Feb 2020 19:50:10 +0000 (19:50 +0000)]
hwpstate_intel(4): Detect and support PKG variant

If package-level control is present, we default to using it.  Per-core
software control may be enabled by setting the machdep.hwpstate_pkg_ctrl
tunable to "0" in loader.conf(5).

4 years agohwpstate_intel(4): Add fallback EPP using PERF_BIAS MSR
cem [Sat, 1 Feb 2020 19:49:13 +0000 (19:49 +0000)]
hwpstate_intel(4): Add fallback EPP using PERF_BIAS MSR

Per Intel SDM (Vol 3b Part 2), if HWP indicates EPP (energy-performance
preference) is not supported, the hardware instead uses the ENERGY_PERF_BIAS
MSR.  In the epp sysctl handler, fall back to that MSR if HWP does not
support EPP and CPUID indicates the ENERGY_PERF_BIAS MSR is supported.

4 years agox86: Add/amend some power-management comments/macros
cem [Sat, 1 Feb 2020 19:46:02 +0000 (19:46 +0000)]
x86: Add/amend some power-management comments/macros

No functional change.

4 years agohwpstate_intel(4): Error check epp sysctl & bail if HW does not support feature
cem [Sat, 1 Feb 2020 19:45:27 +0000 (19:45 +0000)]
hwpstate_intel(4): Error check epp sysctl & bail if HW does not support feature

4 years agotests: epair: Don't fail if we load if_epair
kp [Sat, 1 Feb 2020 19:40:11 +0000 (19:40 +0000)]
tests: epair: Don't fail if we load if_epair

kldload() returns a positive integer when it loads a ko, so check that the
return value is -1 to detect error cases, not that it's different from zero.

MFC after: 3 days
X-MFC-With: r357234

4 years agoRemove a couple of lingering usages of the page lock.
markj [Sat, 1 Feb 2020 18:23:51 +0000 (18:23 +0000)]
Remove a couple of lingering usages of the page lock.

Update vm_page_scan_contig() and vm_page_reclaim_run() to stop using
vm_page_change_lock().  It has no use after r356157.  Remove
vm_page_change_lock() now that it has no users.

Remove an unncessary check for wirings in vm_page_scan_contig(), which
was previously checking twice.  The check is racy until
vm_page_reclaim_run() ensures that the page is unmapped, so one check is
sufficient.

Reviewed by: jeff, kib (previous versions)
Tested by: pho (previous version)
Differential Revision: https://reviews.freebsd.org/D23279

4 years agointel_hwpstate(4): Use identcpu-cached cpuid 6 leaf
cem [Sat, 1 Feb 2020 17:54:46 +0000 (17:54 +0000)]
intel_hwpstate(4): Use identcpu-cached cpuid 6 leaf

No functional change.

4 years agointel_hwpstate(4): Don't leak bound thread in error conditions
cem [Sat, 1 Feb 2020 17:30:45 +0000 (17:30 +0000)]
intel_hwpstate(4): Don't leak bound thread in error conditions

I don't know why a Skylake CPU with the HWP feature bit present would trap
on MSR reads of the HWP registers, but if this occurs, do not leave the
attach thread bound.  This could conceivably cause reported hangs, although
I have no evidence that this is the cause.

Reported by: ae@, Andreas Nilsson <andrnils AT gmail.com>
X-MFC-With: r357002

4 years agoprci: register tlclk as a fixed clock
mhorne [Sat, 1 Feb 2020 17:13:52 +0000 (17:13 +0000)]
prci: register tlclk as a fixed clock

The PRCI exports tlclk as a constant fixed divisor clock, defined as 1/2
of the coreclk frequency. In older FU540 device trees (such as the one
provided by SiFive), tlclk is represented as its own entity, and is
automatically registered as a fixed-divisor-clock. Unfortunately the
upstream FU540 device tree (that we have in our tree) represents tlclk
as an output of the PRCI block, and we must register it manually. At
worst, users of the old device tree will end up with an unreferenced
duplicate of tlclk.

This fixes device attachment for the SiFive UART on newer device trees,
since it references tlclk via the PRCI.

Reviewed by: kp
Differential Revision: https://reviews.freebsd.org/D23406

4 years agoprci: fix up compat
mhorne [Sat, 1 Feb 2020 17:12:15 +0000 (17:12 +0000)]
prci: fix up compat

Add two additional compat strings that can be used to identify the PRCI.

With newer device trees the PRCI has two parents, hfclk and rtcclk, so
allow the driver to attach when more than one parent is found.

Reviewed by: kp
Differential Revision: https://reviews.freebsd.org/D23405

4 years agoprci: register the DDR and GEMGX PLLs
mhorne [Sat, 1 Feb 2020 17:09:56 +0000 (17:09 +0000)]
prci: register the DDR and GEMGX PLLs

The PRCI module exports three PLLs. Currently only the coreclk/corepll
is registered, so add the logic to register the DDR (memory) and GEMGX
(ethernet) clocks as well. These clocks are unused at the moment.

Reviewed by: kp
Differential Revision: https://reviews.freebsd.org/D23404

4 years agoFix new clang 10.0.0 warnings about converting the result of shift
dim [Sat, 1 Feb 2020 17:02:26 +0000 (17:02 +0000)]
Fix new clang 10.0.0 warnings about converting the result of shift
operations to a boolean in tpm(4):

  sys/dev/tpm/tpm_crb.c:301:32: error: converting the result of '<<' to a boolean; did you mean '(1 << (0)) != 0'? [-Werror,-Wint-in-bool-context]
  WR4(sc, TPM_CRB_CTRL_CANCEL, !TPM_CRB_CTRL_CANCEL_CMD);
^
  sys/dev/tpm/tpm_crb.c:73:34: note: expanded from macro 'TPM_CRB_CTRL_CANCEL_CMD'
  #define TPM_CRB_CTRL_CANCEL_CMD         BIT(0)
  ^
  sys/dev/tpm/tpm20.h:60:19: note: expanded from macro 'BIT'
  #define BIT(x) (1 << (x))
    ^

In this case, the intent was to clear the zeroth bit, and leave the rest
unaffected.  Therefore, the ~ operator should be used instead.

Noticed by: cem
MFC after: 3 days

4 years agoRevert r357349, since the clang 10.0.0 warning was actually correct, and
dim [Sat, 1 Feb 2020 16:57:04 +0000 (16:57 +0000)]
Revert r357349, since the clang 10.0.0 warning was actually correct, and
the ! operator should have been a ~ instead:

  Merge r357348 from the clang 10.0.0 import branch:

  Disable new clang 10.0.0 warnings about converting the result of
  shift operations to a boolean in tpm(4):

  sys/dev/tpm/tpm_crb.c:301:32: error: converting the result of '<<' to a boolean; did you mean '(1 << (0)) != 0'? [-Werror,-Wint-in-bool-context]
  WR4(sc, TPM_CRB_CTRL_CANCEL, !TPM_CRB_CTRL_CANCEL_CMD);
^
  sys/dev/tpm/tpm_crb.c:73:34: note: expanded from macro 'TPM_CRB_CTRL_CANCEL_CMD'
  #define TPM_CRB_CTRL_CANCEL_CMD         BIT(0)
  ^
  sys/dev/tpm/tpm20.h:60:19: note: expanded from macro 'BIT'
  #define BIT(x) (1 << (x))
    ^

  Such warnings can be useful in C++ contexts, but not so much in kernel
  drivers, where this type of bit twiddling is commonplace.  So disable
  it for this case.

Noticed by: cem
MFC after: 3 days

4 years agotests: epair: Don't fail if the if_epair module is already loaded
kp [Sat, 1 Feb 2020 11:31:45 +0000 (11:31 +0000)]
tests: epair: Don't fail if the if_epair module is already loaded

kldload() returns an error (EEXIST) if the module is already loaded.
That's not a problem for us, so ignore that error.

While here also clean up include statements.

MFC after: 3 days
X-MFC-With: r357234

4 years agobridge.4: Remove notes about FreeBSD 6.2-7.2
0mp [Sat, 1 Feb 2020 10:25:13 +0000 (10:25 +0000)]
bridge.4: Remove notes about FreeBSD 6.2-7.2

Reported by: Mateusz Kwiatkowski
Reviewed by: brueffer
Approved by: bcr (mentor)
Differential Revision:  https://reviews.freebsd.org/D23393

4 years agoThe error variable is not really needed. Remove it.
pjd [Sat, 1 Feb 2020 10:15:23 +0000 (10:15 +0000)]
The error variable is not really needed. Remove it.

4 years agoRestore previous usage presentation (without "pwait: " prefix).
pjd [Sat, 1 Feb 2020 09:13:11 +0000 (09:13 +0000)]
Restore previous usage presentation (without "pwait: " prefix).

Pointed out by: jilles

4 years agovfs: replace VOP_MARKATIME with VOP_MMAPPED
mjg [Sat, 1 Feb 2020 06:46:55 +0000 (06:46 +0000)]
vfs: replace VOP_MARKATIME with VOP_MMAPPED

The routine is only provided by ufs and is only used on mmap and exec.

Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D23422

4 years agoufs: drop ufs_markatime from ufs_fifoops
mjg [Sat, 1 Feb 2020 06:41:44 +0000 (06:41 +0000)]
ufs: drop ufs_markatime from ufs_fifoops

The routine is only called on mmap and exec, both of which are invalid for
this type.

Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D23421

4 years agovfs: save on atomics on the root vnode for absolute lookups
mjg [Sat, 1 Feb 2020 06:40:35 +0000 (06:40 +0000)]
vfs: save on atomics on the root vnode for absolute lookups

There are 2 back-to-back atomics on the vnode, but we can check upfront if one
is sufficient. Similarly we can handle relative lookups where current working
directory == root directory.

Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D23427

4 years agovfs: add vrefactn
mjg [Sat, 1 Feb 2020 06:39:49 +0000 (06:39 +0000)]
vfs: add vrefactn

Differential Revision: https://reviews.freebsd.org/D23427

4 years agozfs: ZFS_WLOCK_TEARDOWN_INACTIVE_WLOCKED -> ZFS_TEARDOWN_INACTIVE_WLOCKED
mjg [Sat, 1 Feb 2020 06:39:10 +0000 (06:39 +0000)]
zfs: ZFS_WLOCK_TEARDOWN_INACTIVE_WLOCKED -> ZFS_TEARDOWN_INACTIVE_WLOCKED

Fix up the argument used in one case as well.

4 years agovlan: Fix panic when vnet jail with a vlan interface is destroyed
kp [Fri, 31 Jan 2020 22:54:44 +0000 (22:54 +0000)]
vlan: Fix panic when vnet jail with a vlan interface is destroyed

During vnet cleanup vnet_if_uninit() checks that no more interfaces remain in
the vnet. Any interface borrowed from another vnet is returned by
vnet_if_return(). Other interfaces (i.e. cloned interfaces) should have been
destroyed by their cloner at this point.

The if_vlan VNET_SYSUNINIT had priority SI_ORDER_FIRST, which means it had
equal priority as vnet_if_uninit(). In other words: it was possible for it to
be called *after* vnet_if_uninit(), which would lead to assertion failures.

Set the priority to SI_ORDER_ANY, like other cloners to ensure that vlan
interfaces are destroyed before we enter vnet_if_uninit().

The sys/net/if_vlan test provoked this.

4 years agoAdd two missing fences with comments describing them. These were found by
jeff [Fri, 31 Jan 2020 22:21:15 +0000 (22:21 +0000)]
Add two missing fences with comments describing them.  These were found by
inspection and after a lengthy discussion with jhb and kib.  They have not
produced test failures.

Don't pointer chase through cpu0's smr.  Use cpu correct smr even when not
in a critical section to reduce the likelihood of false sharing.

4 years agomake all is needed to generate .depend.*
bdrewery [Fri, 31 Jan 2020 21:08:33 +0000 (21:08 +0000)]
make all is needed to generate .depend.*

PR: 241746
X-MFC-With: r357043
MFC after: 1 week

4 years agosmrstress: Add 'publishing' fences to operations on smrs_current.
kib [Fri, 31 Jan 2020 20:30:50 +0000 (20:30 +0000)]
smrstress: Add 'publishing' fences to operations on smrs_current.

Reported and tested by: andrew
Reviewed by: jeff
Sponsored by: The FreeBSD Foundation
Differential revision: https://reviews.freebsd.org/D23440

4 years agoMerge r357348 from the clang 10.0.0 import branch:
dim [Fri, 31 Jan 2020 19:36:14 +0000 (19:36 +0000)]
Merge r357348 from the clang 10.0.0 import branch:

Disable new clang 10.0.0 warnings about converting the result of shift
operations to a boolean in tpm(4):

sys/dev/tpm/tpm_crb.c:301:32: error: converting the result of '<<' to a boolean; did you mean '(1 << (0)) != 0'? [-Werror,-Wint-in-bool-context]
        WR4(sc, TPM_CRB_CTRL_CANCEL, !TPM_CRB_CTRL_CANCEL_CMD);
                                      ^
sys/dev/tpm/tpm_crb.c:73:34: note: expanded from macro 'TPM_CRB_CTRL_CANCEL_CMD'
#define TPM_CRB_CTRL_CANCEL_CMD         BIT(0)
                                        ^
sys/dev/tpm/tpm20.h:60:19: note: expanded from macro 'BIT'
#define BIT(x) (1 << (x))
                  ^

Such warnings can be useful in C++ contexts, but not so much in kernel
drivers, where this type of bit twiddling is commonplace.  So disable it
for this case.

MFC after: 3 days

4 years agoMerge r357345 from the clang1000-import branch:
dim [Fri, 31 Jan 2020 19:06:49 +0000 (19:06 +0000)]
Merge r357345 from the clang1000-import branch:

Disable new clang 10.0.0 warnings about misleading indentation in flex.

As this is contributed code with very messy indentation, which will
almost certainly never be upgraded, just disable the warning.

MFC after: 3 days

4 years agoMerge r357342 from the clang1000-import branch:
dim [Fri, 31 Jan 2020 19:06:01 +0000 (19:06 +0000)]
Merge r357342 from the clang1000-import branch:

Work around two -Werror warning issues in googletest, which have been
solved upstream in the mean time.

The first issue is because one of googletest's generated headers contain
classes with a user-declared copy assignment operator, but rely on the
generation by the compiler of an implicit copy constructor, which is now
deprecated:

/usr/obj/usr/src/amd64.amd64/tmp/usr/include/private/gtest/internal/gtest-param-util-generated.h:5284:8: error: definition of implicit copy constructor for 'CartesianProductHolder3<testing::internal::ParamGenerator<bool>, testing::internal::ValueArray3<int, int, int>, testing::internal::ValueArray4<cache_mode, cache_mode, cache_mode, cache_mode> >' is deprecated because it has a user-declared copy assignment operator [-Werror,-Wdeprecated-copy]
  void operator=(const CartesianProductHolder3& other);
       ^
/usr/obj/usr/src/amd64.amd64/tmp/usr/include/private/gtest/gtest-param-test.h:1277:10: note: in implicit copy constructor for 'testing::internal::CartesianProductHolder3<testing::internal::ParamGenerator<bool>, testing::internal::ValueArray3<int, int, int>, testing::internal::ValueArray4<cache_mode, cache_mode, cache_mode, cache_mode> >' first required here
  return internal::CartesianProductHolder3<Generator1, Generator2, Generator3>(
         ^
/usr/src/tests/sys/fs/fusefs/io.cc:534:2: note: in instantiation of function template specialization 'testing::Combine<testing::internal::ParamGenerator<bool>, testing::internal::ValueArray3<int, int, int>, testing::internal::ValueArray4<cache_mode, cache_mode, cache_mode, cache_mode> >' requested here
        Combine(Bool(),                                 /* async read */
        ^

For now, silence the warning using -Wno-deprecated-copy.

The second issue is because one of the googlemock test programs attempts
to use "unsigned wchar_t" and "signed wchar_t", which are non-standard
and at best, hazily defined:

contrib/googletest/googlemock/test/gmock-actions_test.cc:111:37: error: 'wchar_t' cannot be signed or unsigned [-Wsigned-unsigned-wchar]
  EXPECT_EQ(0U, BuiltInDefaultValue<unsigned wchar_t>::Get());
                                    ^
contrib/googletest/googlemock/test/gmock-actions_test.cc:112:36: error: 'wchar_t' cannot be signed or unsigned [-Wsigned-unsigned-wchar]
  EXPECT_EQ(0, BuiltInDefaultValue<signed wchar_t>::Get());
                                   ^

For now, silence the warning using -Wno-signed-unsigned-wchar.

MFC after: 3 days

4 years agoAdd stricter checks on user changes to SSTATUS.
jhb [Fri, 31 Jan 2020 19:00:48 +0000 (19:00 +0000)]
Add stricter checks on user changes to SSTATUS.

Rather than trying to blacklist which bits userland can't write to via
sigreturn() or setcontext(), only permit changes to whitelisted bits.

- Permit arbitrary writes to bits in the user-writable USTATUS
  register that shadows SSTATUS.

- Ignore changes in write-only bits maintained by the CPU.

- Ignore the user-supplied value of the FS field used to track
  floating point state and instead set it to a value matching the
  actions taken by set_fpcontext().

Discussed with: mhorne
MFC after: 2 weeks
Sponsored by: DARPA
Differential Revision: https://reviews.freebsd.org/D23338

4 years agoservices: Add PROFInet and EtherCAT.
pfg [Fri, 31 Jan 2020 18:55:21 +0000 (18:55 +0000)]
services: Add PROFInet and EtherCAT.

Both are used in industrial networks.

MFC after: 1 week

4 years agoregen src.conf.5 after r357338 BSD_CRTBEGIN retirement
emaste [Fri, 31 Jan 2020 18:26:13 +0000 (18:26 +0000)]
regen src.conf.5 after r357338 BSD_CRTBEGIN retirement

4 years agoMerge r357339 from the clang1000-import branch:
dim [Fri, 31 Jan 2020 18:13:00 +0000 (18:13 +0000)]
Merge r357339 from the clang1000-import branch:

Fix the following -Werror warning from clang 10.0.0 in bsnmpd:

usr.sbin/bsnmpd/modules/snmp_pf/pf_snmp.c:1661:4: error: misleading indentation; statement is not part of the previous 'else' [-Werror,-Wmisleading-indentation]
                        return (-1);
                        ^
usr.sbin/bsnmpd/modules/snmp_pf/pf_snmp.c:1658:5: note: previous statement is here
                } else
                  ^

The intent was to group the return statement with the previous syslog()
call.

MFC after: 3 days

4 years agoretire BSD_CRTBEGIN option
emaste [Fri, 31 Jan 2020 18:04:04 +0000 (18:04 +0000)]
retire BSD_CRTBEGIN option

BSD crt is currently used on all architectures (other than sparc64).
Remove the option and use BSD crt everywhere as part of the GCC 4.2.1
retirement plan.

https://lists.freebsd.org/pipermail/freebsd-arch/2020-January/019823.html

PR: 239851
Reviewed by: andrew, brooks
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D23122

4 years agoFix 64-bit value of SSTATUS_SD to use an unsigned long.
jhb [Fri, 31 Jan 2020 17:49:15 +0000 (17:49 +0000)]
Fix 64-bit value of SSTATUS_SD to use an unsigned long.

While here, fix MSTATUS_SD to match SSTATUS_SD.

Reviewed by: mhorne
MFC after: 2 weeks
Sponsored by: DARPA
Differential Revision: https://reviews.freebsd.org/D23434

4 years agohwpstate(4): Ignore CurPstateLimit by default
cem [Fri, 31 Jan 2020 17:40:41 +0000 (17:40 +0000)]
hwpstate(4): Ignore CurPstateLimit by default

Add a sysctl knob to allow users to re-enable it, and document the knob and
default in cpufreq.4.  (While here, add a few unrelated updates to
cpufreq.4.)

It seems that the register value in some hardware simply reflects the
configured P-state.  This results in an inadvertent and unintended outcome
where the P-state can only walk down, and then the driver becomes "stuck" in
the slowest possible P-state.

The Linux driver never consults this register, so that's some evidence that
ignoring the contents are relatively harmless.

PR: 234733
Reported by: sigsys AT gmail.com, Erich Dollanksy <freebsd.ed.lists AT
sumeritec.com>

4 years agoRPI3: Add RPi4 firmware files to the FAT partition
kevans [Fri, 31 Jan 2020 15:56:08 +0000 (15:56 +0000)]
RPI3: Add RPi4 firmware files to the FAT partition

I've discovered I have this local diff that never got committed -- this
should have been a part of r355424.

Reproted by: Klaus Küchemann <maciphone2@googlemail.com>

4 years agoReimplement stack capture of running threads on i386 and amd64.
markj [Fri, 31 Jan 2020 15:43:33 +0000 (15:43 +0000)]
Reimplement stack capture of running threads on i386 and amd64.

After r355784 the td_oncpu field is no longer synchronized by the thread
lock, so the stack capture interrupt cannot be delievered precisely.
Fix this using a loop which drops the thread lock and restarts if the
wrong thread was sampled from the stack capture interrupt handler.

Change the implementation to use a regular interrupt instead of an NMI.
Now that we drop the thread lock, there is no advantage to the latter.

Simplify the KPIs.  Remove stack_save_td_running() and add a return
value to stack_save_td().  On platforms that do not support stack
capture of running threads, stack_save_td() returns EOPNOTSUPP.  If the
target thread is running in user mode, stack_save_td() returns EBUSY.

Reviewed by: kib
Reported by: mjg, pho
Tested by: pho
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D23355

4 years agoCall the MAPTI command earlier in the ITS driver
andrew [Fri, 31 Jan 2020 11:33:11 +0000 (11:33 +0000)]
Call the MAPTI command earlier in the ITS driver

The GICv3 Software Overview suggests when allocating a new MSI/MSI-X
interrupt we need to call MAPD followed by MAPTI. Unfortunately the code
would place a MOVI command between these. This is invalid as it needs
values set by the MAPTI to be present.

Re-order so we allocate a temporary CPU for the interrupt, then use the
MAPTI command to assign the MSI to it.

MFC after: 2 weeks
Sponsored by: DARPA, AFRL

4 years agovfs: revert the overzealous assert added in r357285 to vgone
mjg [Fri, 31 Jan 2020 11:31:14 +0000 (11:31 +0000)]
vfs: revert the overzealous assert added in r357285 to vgone

The intent was to make it more likely to catch filesystems with custom
need_inactive routines which fail to call vn_need_pageq_flush (or do an
equivalent).

One immediate case which is missed is vgone from called by inactive itself.

A better assertion may land later. The routine is not added to vputx because
it is of no use to tmpfs et al.

Reported by: syzbot+5f697ec11f89b60941db@syzkaller.appspotmail.com

4 years agoRevert r357293.
hselasky [Fri, 31 Jan 2020 10:51:13 +0000 (10:51 +0000)]
Revert r357293.
The netisr uses rm_ locks not rms_ locks as noted by jeff@ .

Sponsored by: Mellanox Technologies

4 years agoWiden EPOCH(9) usage in mlx4en(4).
hselasky [Fri, 31 Jan 2020 10:41:47 +0000 (10:41 +0000)]
Widen EPOCH(9) usage in mlx4en(4).

Make sure all receive completion callbacks are covered by the network
EPOCH(9), because this is required when calling if_input() and
ether_input() after r357012.

Convert some spaces to tabs while at it.

Sponsored by: Mellanox Technologies