MI linux.[c|h] are the module independent in terms of the Linux emulation
layer (ie, intended for both ISA - 32 & 64 bit), analogue of MD linux.h.
There must be a code here that cannot be placed into the corresponding by
common sense MI source and header files, i.e., code is machine independent,
but ISA dependent.
For the use_real_names knob, the code must be placed into the
linux_socket.[c|h], however linux_socket is ISA dependent.
MFC after: 2 weeks
This obsolete system call is not supported by glibc. In ancient libc
versions (before glibc 2.0), uselib() was used to load the shared
libraries with names found in an array of names in the binary.
On Linux, since 3.15, this system call is available only when
the kernel is configured with the CONFIG_USELIB option.
It doesn't look like anyone needs this syscall for others Linuxulators,
so move it to the corresponding MD Linuxulator.
MFC after: 2 weeks
Include sys/sysent.h directly where it needed. The linux_util.h included
in a most source files of the Linuxulator, avoid collecting a rarely used
includes here.
MFC after: 2 weeks
Include vm headers directly where they needed. The linux_util.h included
in a most source files of the Linuxulator, avoid collecting a rarely used
includes here.
MFC after: 2 weeks
`cpu_data(cpu)` evaluates to a `struct cpuinfo_x86` filled with
attributes of the given CPU number. The CPU number is an index in the
`__cpu_data[]` array with MAXCPU entries. On FreeBSD, we simply
initialize all of them like we do with `boot_cpu_data`.
While here, we add the `x86_model` field to the `struct cpuinfo_x86`. We
use `CPUID_TO_MODEL()` to set it.
At the same time, we fix the value of `x86` which should have been set
to the CPU family. It was using the same implementation as
`CPUID_TO_MODEL()` before. It now uses `CPUID_TO_FAMILY()`.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38542
`xa_is_err()` is synonymous to `IS_ERR()`.
Other introduced functions call their equivalent without the `irq*`
suffix.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38534
For `vga_client_register()`, the API is modified twice in a row. To keep
the API compatible with all commits in the DRM driver, we introduce two
`LINUXKPI_VERSION` version bumps.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38533
This is not used by the DRM driver yet because we comment out the code
calling them, but they are easy to implement.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38530
Following 5a9a0d7803 initialize the
queue values explicitly and deal with a stopped queue in
ieee80211_tx_dequeue().
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
amdgpu's virtual display feature uses pci_name() to match a module parameter
string, and the documentation shows an example of `0000:26:00.0` for the name.
In our case the name was just `drmn`, which is not actually unique across
devices.
The other consumers are wireless drivers, which will benefit from this
change.
Generate the expected string for pci_name() to return.
Related to: https://github.com/freebsd/drm-kmod/issues/134
Sponsored by: https://www.patreon.com/valpackett
Reviewed by: bz, hselasky, manu (earlier)
MFC after: 3 days
Differential Revision: https://reviews.freebsd.org/D34248
Following e9715b1c44 and
4b56afaf7b, implement support
for up-to 32 MSI vectors. This is used by wireless drivers.
This also switches msi_desc to an array in order to store
per-vector information.
Sponsored by: The FreeBSD Foundation
Discussed with: grehan (in Dec)
MFC after: 3 days
Reviewed by: jhb
Differential Revision: https://reviews.freebsd.org/D38222
irq_set_affinity_hint() takes a const mask argument and some drivers
pass it in as such where earlier implementations were more lenient.
Deal with it and __DECONST() the argument when passed to intr_setaffinity().
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
Differential Revision: https://reviews.freebsd.org/D38242
Add a dummy irq_set_status_flags() along with #defines passed by the driver.
Add disable_irq_nosync() as another wrapper to lkpi_disable_irq().
Those are used by wireless drivers.
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
Differential Revision: https://reviews.freebsd.org/D38241
Add a dummy implementation of pm_wakeup_event() which is used to notify
the power management system about a wakeup (which we currently do not
implement yet).
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
Differential Revision: https://reviews.freebsd.org/D38239
Add a dummy device_set_wakeup_enable() which is used for WoWLAN which we
do not (yet) support and device_wakeup_enable() which is a wrapper to the
former with the enable argument being true.
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
Differential Revision: https://reviews.freebsd.org/D38238
Add more empty header files wireless drivers try to include but
we do not (yet) need for any implementation.
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
Differential Revision: https://reviews.freebsd.org/D38271
Remove former dummy/ files which now have an actual implementation
in common/ .
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
Differential Revision: https://reviews.freebsd.org/D38269
Add a dummy pci_assign_resource() and an implementation of
pci_irq_vector() returning the irq for MSI-X, MSI, and legacy interrupt.
Both are needed by wirless drivers.
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
Reviewed by: jhb
Differential Revision: https://reviews.freebsd.org/D38237
Summary:
Migrate linprocfs to use the IfAPI interfaces instead of direct ifnet
accesses.
Reviewed by: dchagin
Sponsored by: Juniper Networks, Inc.
Differential Revision: https://reviews.freebsd.org/D38358
COMPAT_LINUX32 option is defined for case when building 32-bit Linuxulator
for the 64-bit host. Usage of __ELF_WORD_SIZE is wrong here as it is equal to 32
on i386 too.
MFC after: 1 week
Add code (currently disabled by #ifdef) for HT and VHT to
lkpi_scan_ies_add(). Switch to a local variable for ic given
the new code also needs the value.
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
This change breaks the API of `list_sort()`. `LINUXKPI_VERSION >= 51300`
is used to keep the header compatible with both versions of the
prototype.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38082
Implement ieee80211_is_pspoll() and move it up to the set of implemented
ieee80211_is_*() functions.
There are no further changes (git diff just looks weird).
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
Very basic implementations of ieee80211_{wake,stop}_queue[s],
as well as ieee80211_txq_schedule_start(), ieee80211_next_txq(),
and ieee80211_schedule_txq().
Various combinations of these are used by different wireless
drivers, incl. iwlwifi.
Sponsored by: The FreeBSD Foundation (parts of this work)
MFC after: 3 days
Add an implementation of ieee80211_tx_status_ext() extending the
internal linuxkpi_ieee80211_tx_status() implementation.
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
Extend linuxkpi_ieee80211_rx() by another argument for
the ieee80211_rx_list() implementation (even though the argument
is currently ignored).
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
Correct struct member types, fix whitespace, and remove superfluous
return statements in void functions.
No functional changes.
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
I'm sure I got it wrong but at least the DRM drivers compile.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38157
`io_mapping_map_local_wc()` is synonymous to `io_mapping_map_atomic_wc()`.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38156
They are synonymous to `kmap_atomic_prot()` and `kunmap_atomic()`
respectively.
While here, fix several style(9) issues.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38155
For now, it's empty. It is included by the DRM drivers but nothing is
missing otherwise. Perhaps something we already defined should be in
`pgtable.h` instead of another header.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38153
For now, it's empty. It is included by the DRM drivers but nothing is
missing otherwise. Perhaps something we already defined should be in
`limits.h` instead of another header.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38147
The struct layout is modified with this commit because new fields are
added in the middle, keeping original Linux order.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D37932
Rework the routines to convert a native statfs structure (with fixed-size 64-bit
counters) to a Linux statfs structure (with long-sized counters) for 32-bit apps.
Instead of following Linux and return an EOVERFLOW error from statfs() family of
syscalls when actual fs stat value(s) are large enough to not fit into 32 bits,
apply scale logics used by FreeBSD to convert a 5.x statfs structure to a 4.x
statfs structure.
For more details see cc479dda.
Tested by: glebius
MFC after: 1 week
They were defined in the i915 DRM driver. I move the code in linuxkpi so
it can benefit other drivers.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38078
It now returns a `struct scatterlist *` pointer instead of an error
code only.
The implementation is incomplete because it doesn't use the `prv`
argument.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38077
At least one file in the DRM drivers benefits from some namespace
pollution to use `fs_reclaim_acquire()`/`fs_reclaim_release()`. They are
defined in `linux/sched/mm.h` and this header must be included
indirectly into the DRM drivers' source file.
I couldn't find how it was included. Therefore this commit includes
`linux/sched/mm.h` from `linux/sched.h`. This is not the case in Linux
but fixes the issue with the DRM drivers.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D37912
This is the same as `debugfs_create_file()` but takes the initial size
of the file. In FreeBSD, the given size is ignored and
`debugfs_create_file()` is called.
Reviewed by: emaste, manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D37914
At the same time, define `PCI_MSIX_FLAGS`and `PCI_MSIX_FLAGS_ENABLE`.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38160
I'm not sure this is correct, but it gives access to the definition of
`PAGE_KERNEL` in `ttm_tt.c` in the DRM drivers.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38154
It is defined as a synonymous to `DEFINE_WW_CLASS(name)`.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38151
`SZ_2K` and `SZ_1G` were defined in irdma. They are removed from this
driver.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38150
This is the same error code as Linux. This value is unused on FreeBSD.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38149
By (ab)using `atomic_long_add_return()`, `atomic_long_sub()` was making
the atomic long overflow. Indeed the underlying FreeBSD atomic is based
on an unsigned long.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38090
The code comes from the i915 DRM driver.
In Linux commits b739f125e4ebd73d10ed30a856574e13649119ed and
b12d691ea5e01db42ccf3b4207e57cb3ce7cfe91 (Linux 5.13), the i915 DRM
driver dropped specific implementations to use Linux generic functions.
Therefore I moved the FreeBSD code from that i915 driver to linuxkpi.
However, these commits were later reverted (also in Linux 5.13) so the
i915 driver doesn't use these functions. But perhaps it will help in the
future.
To sum up, the code comes from the i915 DRM driver but it doesn't use it
(i.e. it continues to use its internal implementation).
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38088
The `debugfs_create_x*()` variants are the same as their
`debugfs_create_u*()` equivalent, but they work with lowercase
hexadecimal.
While here, fix a few style(9) issues.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D38087
It is the same as callout_stop(9) but the return values are different.
Reviewed by: hselasky
Approved by: hselasky
Differential Revision: https://reviews.freebsd.org/D38081
This is the same error code as Linux.
As emaste@ noted in the review, FreeBSD defines the following errno
values in `sys/errno.h`:
* 56 is `EISCONN`
* 57 is `ENOTCONN`
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D37935
It uses the `VM_MEMATTR_WRITE_BACK` flag on FreeBSD.
It replaces `ioremap_wb()` which doesn't exist in Linux. Perhaps it
existed in the past and was removed.
Reviewed by: emaste, manu
Approved by: emaste, manu
Differential Revision: https://reviews.freebsd.org/D37916
bz@ asked if the KBI breakage is a concern here. My answer was that this
is the first time in the DRM drivers in Linux 5.13 (the version I'm
working on) that this structure is initialized (as a variable local to
the function in this case), so it shouldn't be a problem for the DRM
drivers.
However, I can't speak for other drivers maintained outside of the src
tree.
Reviewed by: emaste, manu
Approved by: emaste, manu
Differential Revision: https://reviews.freebsd.org/D37913
Add irq_get_msi_desc() as a wrapper around a PCI function which will
allocate a single cached value (see comment on struct) for the
msi_desc requested if it doesn't exist yet and handle freeing it
when the PCI device goes away. We take the values from the ivars of
the native (FreeBSD) device.
While changing struct pci_dev also add the msi_cap field requested by
a wireless driver.
Bump __FreeBSD_version so these changes can be detected.
MFC after: 3 days
X-MFC: move fields to end of struct (alloc happens in linux_pci.c)
Reviewed by: hselasky (earlier version)
Differential Revision: https://reviews.freebsd.org/D37523
Add a version of pci_get_device() as linuxkpi_pci_get_device()
not (yet) supporting the last argument.
Due to conflicts we cannot redefine it as we would normally do
in LinuxKPI so drivers have to be adjusted.
MFC after: 3 days
Differential Revision: https://reviews.freebsd.org/D37593
pci_alloc_irq_vectors() is given a min and max vector value.
pci_enable_msi() will always succeed independent of these arguments as
it does not know about them. Further it will only ever allocate
1 "vector" not supporting any other amount.
So upfront check that (a) the available pci_msi_count() can satisfy the
requested minv and (b) given the pci_enable_msi() hard coded limit check
that minv is not larger than 1.
If we cannot satisfy either requirement return an error.
This fixes problems with drivers which check that the returned value
of allocated "vectors" will match their requests and only otherwise try
to fall back to ask for 1 or deal otherwise.
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
Reviewed by: hselasky (earlier version)
Differential Revision: https://reviews.freebsd.org/D37522
Add more functions to netdevice.h (netif_napi_add_tx() being the only
one implemented) and add platform_device.h and netlink.h in order to
make driver code compile.
The skeleton functions are used only in very limited scope and not at
all in our usage so far but add (invasive) #ifdef if removed.
Add pr_debug() calls to each of them in order to log a TODO (if DEBUG
compiled in) and someone should hit them in the future.
MFC after: 3 days
Commented on by: hselasky (earlier version)
Differential Revision: https://reviews.freebsd.org/D37599
While here:
- fix an argument of kstrtouint_from_user() to correct signedness.
- make kstrtou32() call kstrtouint() to avoid duplication (keep inline
function)
Add kstrtou32_from_user() based on other examples in the file
making it a copy of the now fixed kstrtouint_from_user().
Also add a rudimentarily hacked up version of mac_pton() which is
leanient accepting non-well-formed input but so far only with ':'
separators. It does not seem to obviously belong to any networking
header file so add it here.
Both new functions are needed for debugfs support for iwlwifi hence
coming together in one commit.
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
Commented on by: emaste
Differential Revision: https://reviews.freebsd.org/D37088
Fix a possible NULL pointer deref in case alloc_pages() fails.
This is theoretical so far as up to now no code in the tree uses
linuxkpi_page_frag_alloc().
Reported by: Coverity via emaste
Coverity ID: 1502345
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
X-MFC-with: 55038a6306
This is defined as a no-op even when INVARIANTS is defined. I admit I
don't know how to implement that in FreeBSD and didn't search
thoroughly.
Reviewed by: bz
Approved by: bz
Differential Revision: https://reviews.freebsd.org/D37911
This defines `ENODEV` used in this header. This fixes a build failure in
the DRM drivers.
Reviewed by: bz
Approved by: bz
Differential Revision: https://reviews.freebsd.org/D37909
Extend pm.h by pm_sleep_ptr and DEFINE_SIMPLE_DEV_PM_OPS().
For the moment this duplicates some parts (as can be seen in the earlier
review I tried to simplify bits but given our implementation this
was easier in the end).
While here and cleanup the SIMPLE_DEV_PM_OPS() bits (white-space only).
MFC after: 3 days
Reviewed by: hselasky
Differential Revision: https://reviews.freebsd.org/D37527
For the moment and the currently only consumer (mt76) add a simplified
version of the page_frag_cache. We will only accept fragement sizes up
to 1 PAGE_SIZE (KASSERT) and we will always return a full page.
Should we add more consumers or small (or large) objects would become a
problem we can always add a more elaborate version.
Discussed with: markj
Reviewed by: markj (,hselasky commented as well)
MFC after: 3 days
Differential Revision: https://reviews.freebsd.org/D37595
Migrate the two functions from the header into the implementation file
in order to have access to more facilities and not to run into possible
allocation/locking/... problems in the future.
Sponsored by: The FreeBSD Foundation
MFC after: 10 days
Implement ieee80211_get_hdrlen_from_skb() doing basic sanity checks
on lengths (minimal length or skb data length vs. header length).
Sponsored by: The FreeBSD Foundation
MFC after: 10 days
Using the previous changes implement cfg80211_get_ies_channel_number()
either based on DSPARMS (or for the future HTINFO).
Sponsored by: The FreeBSD Foundation
MFC after: 10 days
Implement cfg80211_find_elem(), ieee80211_bss_get_elem(),
ieee80211_bss_get_ie(), and cfg80211_find_vendor_ie() with the last
one having a short cut always also checking oui_type in the pattern.
Sponsored by: The FreeBSD Foundation
MFC after: 10 days
Implement cfg80211_{get,put}_bss currently doing malloc/free bits,
so hopefully the drivers get the calls right.
cfg80211_get_bss() sets up a lookup structure which may also take a
result (first hit wins) and calls ieee80211_scan_iterate() comparing
the various values in the iterator funcion. Some of the checks are
partially pointless (as it seems the drivers are not interested in
these parts [ANY] but we keep them for documentation purposes should
futher values arise in the future).
We currently only iterate over the first VAP which will do for now.
Sponsored by: The FreeBSD Foundation
MFC after: 10 days
When initializing the sta set the per-band supported legacy rates
as some drivers take the information from there.
Sponsored by: The FreeBSD Foundation
MFC after: 10 days
in seqcount_mutex_t case with removal of extraneous mutex lock/unlock
calls and addition of missing critical section.
While here strip one inline wrap layer to reduce code size.
Fixes startup lockup of i915kms after update to drm-kmod v5.12
Reviewed by: hselasky, bz
MFC after: 1week
Differential Revision: https://reviews.freebsd.org/D37699
This unbreak drm-kmod build.
the const is part of Linux API
Unfortunately drm-kmod uses hand-rolled refcount* calls on a kref
object. For now go the easy route of keeping it operational by casting
stuff internally.
The general goal here is to make FreeBSD refcount API use an opaque
type, hence the ongoing removal of hand-rolled accesses.
Reported by: emaste
dma_map_sgtable internally uses the dma_map_sg_attrs helper. The problem is
that dma_map_sg_attrs returns the number of entries mapped, whereas
dma_map_sgtable returns nonzero on failure. This leads to dma_map_sgtable
returning non-zero-but-positive values which tricks other areas of the stack
into thinking nents is a valid pointer.
This checks if nents is valid and returns zero if so, updating the nents field
in sgt. This fixes PRIME render offload with nvidia-drm.
Fixes: 9202c95f47 ("linuxkpi: Add dma_{un,}map_sgtable")
In Linux, this limits the accepted value to -1, 0 and 1.
In FreeBSD, this remains a signed integer with no specific constraints.
This change is a requirement to update our DRM drivers to Linux 5.12.
Differential Revision: https://reviews.freebsd.org/D37364
To achieve that, the header uses the C11 type generic selection keyboard
_Generic() because the macros are supposed to work with seqcount_t
and seqcount_mutex_t.
Differential Revision: https://reviews.freebsd.org/D36965
- add comments for enum values constantly looked up, and another one to
a net80211 equivalent (should possibly re-define those in the future?)
- add another nl80211_sta_info flag
- add enum environment_cap used in cfg80211.h in the future.
MFC after: 3 days
- skb_reset_tail_pointer(): we do not do offsets so do a plain reset
- skb_add_rx_frag(): adjust data_len to keep track of the frag
- based on that implement skb_is_nonlinear() and skb_linearize()
- implement build_skb() and adjust linuxkpi_kfree_skb() and ddb macro.
Sponsored by: The FreeBSD Foundation (partially)
MFC after: 3 days
In preparation for future updates remove the budget argument from the
netif_napi_add() in drivers and update LinuxKPI to reflect that it is
gone and only set it internally. This required changes to the currently
committed wireless drivers based on LinuxKPI (iwlwifi, rtw88, rtw89).
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
While we do not currently use ethtool, add the definitions to avoid
other longer-term maintenance problems with drivers.
Also migrate ETH_GSTRING_LEN into here from if_ether.h as it seems this
is where it belongs.
MFC after: 3 days
Differential Revision: https://reviews.freebsd.org/D37214
Add debugfs_create_u8() based on other already present implementations.
Add a read-only implementation for debugfs_create_blob().
Both are needed for iwlwifi debugfs support.
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
OKed by: jfree (earlier version)
Differential Revision: https://reviews.freebsd.org/D37090
This is needed for debugfs implementations in drivers.
Sponsored by: The FreeBSD Foundation
MFC after: 3 days
OKed by: jfree
Differential Revision: https://reviews.freebsd.org/D37092
Add more MSI related constansts defined to our native defines and
pci_is_enabled(). All are needed for another wireless driver.
MFC after: 3 days
Differential Revision: https://reviews.freebsd.org/D37225
Add a memset_startat() macro which sets a pattern from a struct member
to the end of the struct. Needed by a wireless driver.
MFC after: 3 days
Reviewed by: emaste
Differential Revision: https://reviews.freebsd.org/D37389
On Linux, the `kmalloc()` family of functions returns a special value if
the size of the allocation is zero. This macro verifies if the pointer
is NULL (the allocation failed) or the size is 0 (the allocation was not
performed AFAIU). This special value can be passed to `kfree()`.
On FreeBSD, our `malloc(9)` functions don't return a special value for
0-size allocations. Therefore we can simply compare the result against
NULL.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D37367
They are not really used in this header. However they are included in
Linux and at least the DRM drivers unfortunately rely on this namespace
pollution.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D37365
This is used by `i915_gem.c` in the i915 DRM driver to get access to
`wbinvd_on_all_cpus()`.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D36970
I took the implementation from OpenBSD, commit
d55ef580b1748517027c3eabdb715316ca5b1442.
The only difference is the addition of `dma_buf_map_is_equal()`.
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D36963
It simply includes the same header in FreeBSD (which is located
elsewhere).
Reviewed by: manu
Approved by: manu
Differential Revision: https://reviews.freebsd.org/D36960