When witness(4) detects lock order reversals (LORs), it prints
information about the stack trace which caused the LOR. If available,
it can also print information about the first stack trace which
established the other lock ordering. However, it only does this for
"simple" LORs where the two locks in question were directly locked
in the opposite order. When the lock order was established through
a more complex pattern of intermediate locks, WITNESS only prints
the stack trace where it detected the LOR.
This commit provides new functionality to provide more verbose
information about the lock chain(s) which established the lock
ordering. The new functionality can be disabled by setting the
debug.witness.trace sysctl/tunable to 1. The new functionality
is also available through the debug.witness.badstacks sysctl,
which has been modified to always show the more verbose
information.
Reviewed by: markj, glebius (previous version), kib (previous version)
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D54785
MFC after: 1 month
Tests should declare their dependencies rather than testing for them at
runtime.
Reviewed by: igoro, kp, glebius
MFC after: 2 weeks
Sponsored by: OPNsense
Sponsored by: Klara, Inc.
Differential Revision: https://reviews.freebsd.org/D54846
If a rule has a divert port set, then we can reasonably predict that
ipdivert.ko is loaded, and in particular that ip_divert_ptr is set.
Moreover, in this case, if ipdivert.ko is not loaded we should just drop
the packet instead of ignoring the divert rule.
Reviewed by: igoro, kp, glebius
MFC after: 2 weeks
Sponsored by: OPNsense
Sponsored by: Klara, Inc.
Differential Revision: https://reviews.freebsd.org/D54845
The previous commit added <sys/cdefs.h>, which isn't actually needed.
Conversely, <sys/event.h> is needed (and has been for a long time) but
was not included.
MFC after: 1 week
Fixes: a11d132f6c ("devstat: Provide 32-bit compatibility")
Reported by: gallatin@
In an MP world, it doesn't hold that x86 requires no memory barriers.
This change should also fix panics due to out-of-sync data seen with
FreeBSD VMs on top of OpenStack and HBAs of type lsiLogic. [1]
While at it:
- Improve the granularity somewhat by distinguishing between read and
write memory barriers as well as refer to existing *mb(9) functions
instead of duplicating these [2], unless IO barriers are also used.
- Nuke the unused SYM_DRIVER_NAME macro.
PR: 270816 [1]
Obtained from: BSD-licensed Linux sym53c8xx driver [2]
MFC after: 1 week
This also fixes sym_cam_free() to tear things down correctly, i. e. in
opposite order of setup, as well as sym_cam_attach() to not free devq
and SIM twice in the failure case.
Set the MK_LLVM_LINK_STATIC_LIBRARIES knob to "yes" when building the
native-xtools target. This reverts to the behaviour prior to
2e47f35be5.
This avoids a build failure that occurs otherwise, where compilation
fails looking for a libllvmprivate.so that was not built.
It is unclear if this addresses the issue in all instances---some
replies in the PRs indicate otherwise. Still, some report success, and
in my own testing this fixed creation of a cross-compiled poudriere
jail. Commit this while we continue to investigate...
PR: 286710, 291409
Tested by: marck, rdunkle@smallcatbrain.com
Reviewed by: emaste
MFC after: 3 days
Fixes: 2e47f35be5 ("Convert libllvm, libclang and liblldb into private shared libraries").
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D54815
Extend pmap_bootstrap_dmap() to build the DMAP with 4K-page granularity.
Recently we have been approximating it with 2MB mappings. The motivation
again is the problematic FU540 hardware, which seems to require more
accurate mappings still to avoid triggering its PMP errata.
Although this hardware alone is of little consequence, constructing the
DMAP accurately/correctly may help avoid future surprises.
The implementation contains some repetitive code. This could be
expressed differently, but my guiding principle for these early routines
is that being simple and explicit about what we are doing makes them
easier to comprehend.
See also 762a3224cd ("riscv: smarter DMAP construction).
Tested by: Klaus Küchemann <maciphone2@googlemail.com>
MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D54716
Add SPMC (system power management controller) driver as acpi_spmc. This
is the device which provides the LPI device D-state constraints and
allows for OSPM to send S0ix/modern standby entry/exit notifications.
This supports the original Intel DSM
(https://uefi.org/sites/default/files/resources/Intel_ACPI_Low_Power_S0_Idle.pdf,
untested), the AMD DSM (tested), and the Microsoft DSM (tested).
Before entry, acpi_spmc_check_constraints is called to notify of any
violated power constraints. This will use acpi_pwr_get_state to get
current device D-states when that gets added back.
Reviewed by: olce
Tested by: jkim, Oleksandr Kryvulia, Matthias Lanter
Approved by: olce
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D48387
If a 32-bit process running on a 64-bit kernel requests kern.devstat.all,
translate each struct devstat to its 32-bit equivalent before copying it
out.
Also fix a bug where an early error would be ignored if there were no
devices to report.
MFC after: 1 week
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D54591
Allow MODULE_DEPEND to accept macros by adding an extra level of
indirection. Some drivers in the tree actually depend on this being the
case.
Reviewed by: imp
Differential Revision: https://reviews.freebsd.org/D54874
Revert the argument change that broke libcam in 8c35de49 and move
power_condition support to scsi_start_stop_pc().
Reported by: imp
Reviewed By: #cam, imp (mentor)
Sponsored by: Samsung Electronics
Differential Revision: https://reviews.freebsd.org/D54822
kqueue1() was meant.
While here, make the wording in the AUTHORS section more clear.
PR: 291908
Fixes: 9b1585384d ("kqueue.2: Editorial pass")
Reviewed by: kib, dab
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D54858
Most of these tests create a md(4) device backed by a makefs-generated
image, then import the pool, using the md device as a pool vdev. When
a test finishes, it destroys the pool, then destroys the md device.
Once in a while md device destruction fails with EBUSY. This happens
when a GEOM consumer is holding the device open; kern_mddetach_locked()
simply fails if any consumers are present. zpool destruction is
synchronous so ZFS is innocent, rather, the problem seems to be that
GEOM re-tastes the device after the zpool reference goes away. This
operation creates a transient reference that causes the spurious
failure.
Use "mdconfig -o force" to work around this.
MFC after: 2 weeks
If execfd is set, the fexecve(2) is used by posix_spawn() instead of the
provided path.
Reviewed by: markj
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D54862
Add <linux/eventfd.h> and expose the `eventfd_*()` API. This is used by
DRM drivers for some time, but the code was commented out so far.
Note that Linux uses `struct eventfd_ctx`, but FreeBSD defines `struct
eventfd`. We define `eventfd_ctx` as a synonym to `eventfd`.
Reviewed by: christos, markj
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D50853
The `eventfd_signal()` function is the equivalent to a write to an
eventfd file descriptor: it bumps the internal counter and wakes up
processes waiting for it.
`eventfd_signal()` is meant to be used by kernel drivers. DRM drivers
will call it through linuxkpi.
Reviewed by: markj
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D50850
An eventfd file descriptor can be used by drivers such as DRM drivers
through linuxkpi. A driver can hold a reference to such a file
regardless of the fact it is used by userland or not.
This patch introduces a refcount in `struct eventfd`, plus the
`eventfd_get()` and `eventfd_put()` functions to acquire and release
references. These functions will be used by DRM drivers for instance.
This structure is internal to `sys/kern/sys_eventfd.c` and not used
anywhere else. Thus it is safe to add a field without breaking anything.
Reviewed by: markj
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D50849
Make pdwait(2) cancellable, same as all other wait*(2) syscalls wrappers.
Reviewed by: asomers, markj
Tested by: pho
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D54592
Jonathan T. Looney