When pledged, if a process receives a bad descriptor the receiver should
not be killed. The EPERM approach used for other conditions is good enough.
Report from Henry Ford
this will be errata 7.8/022_recvfd.patch and 7.7/028_recvfd.patch
drm/sched: Fix kernel-doc warning for drm_sched_job_done()
From Yujie Liu
da09dfc90cb7ed1ab40d675234382f151eeb0563 in linux-6.18.y/6.18.17
61ded1083b264ff67ca8c2de822c66b6febaf9a8 in mainline linux
drm/syncobj: Fix handle <-> fd ioctls with dirty stack
From Julian Orth
7196a1ff7b9a2ab6d973fe3c1dfc426d8d8ed4d2 in linux-6.18.y/6.18.17
2e3649e237237258a08d75afef96648dd2b379f7 in mainline linux
drm/amd/display: Use GFP_ATOMIC in dc_create_stream_for_sink
From Natalie Vock
0381584929791c4b989fb0a36a466ae20aea1608 in linux-6.18.y/6.18.17
28dfe4317541e57fe52f9a290394cd29c348228b in mainline linux
drm/i915/dp: Fix pipe BPP clamping due to HDR
From Imre Deak
9498fa25a0b0d8c095ce3d1f15d7864228692822 in linux-6.18.y/6.18.17
fe26ae6ac8b88fcdac5036b557c129a17fe520d2 in mainline linux
drm/i915/dp: Fail state computation for invalid DSC source input BPP values
From Imre Deak
99f617ea2ff017b0ba10d5371d83345331091afa in linux-6.18.y/6.18.17
338465490cf7bd4a700ecd33e4855fee4622fa5f in mainline linux
drm/amd: Fix hang on amdgpu unload by using pci_dev_is_disconnected()
From Mario Limonciello
378dff71efddd15f34124bf9d7c98cd69cd05286 in linux-6.18.y/6.18.17
f7afda7fcd169a9168695247d07ad94cf7b9798f in mainline linux
drm/amdgpu: Fix error handling in slot reset
From Lijo Lazar
73e8bdf14248136459753252a438177df7ed8c7c in linux-6.18.y/6.18.17
b57c4ec98c17789136a4db948aec6daadceb5024 in mainline linux
drm/amdgpu: Fix locking bugs in error paths
From Bart Van Assche
5c42d9f8b004905eb441f0eab34662bdacf3192f in linux-6.18.y/6.18.17
480ad5f6ead4a47b969aab6618573cd6822bb6a4 in mainline linux
drm/amdgpu: Unlock a mutex before destroying it
From Bart Van Assche
39ae59e5c1d880176c41ecc2694e0ae81335079d in linux-6.18.y/6.18.17
5e0bcc7b88bcd081aaae6f481b10d9ab294fcb69 in mainline linux
drm/amdgpu/userq: Do not allow userspace to trivially triger kernel warnings
From Tvrtko Ursulin
1753f5f81ab60a553287f9ee659a6ac363adf8d7 in linux-6.18.y/6.18.17
7b7d7693a55d606d700beb9549c9f7f0e5d9c24f in mainline linux
drm/client: Do not destroy NULL modes
From Jonathan Cavitt
4e3ca5f82346cc23c0a71f1ceb006115ff6b0745 in linux-6.18.y/6.18.17
c601fd5414315fc515f746b499110e46272e7243 in mainline linux
drm/fourcc: fix plane order for 10/12/16-bit YCbCr formats
From Simon Ser
cd2713a37336d839027f60ecb3964f1c595991bf in linux-6.18.y/6.18.17
e9e0b48cd15b46dcb2bbc165f6b0fee698b855d6 in mainline linux
xargs: Make the -0 option override -E.
POSIX 1003.1-2024 states that -0 and -E are mutually exclusive, but
leaves unspecified the actual behavior when both are specified.
Unlike GNU xargs, we don't emit a warning when both -0 and -E are
given. OK daniel@
Fix unveil to handle a mount point on a mount point
When traversing upward to find a covering unveil, we always
looked to ascend in the mount point filesystem when we encounter
the root of a mounted filesystem.
We missed the case where the mount point was also the root
of another mounted filesystem, causing us to then fail to ascend,
potentially not finding an unveil above that point.
Found by naddy by mounting an mfs /usr/obj on /usr/obj which
was also a mounted filesystem, which then missed the unveil
in "/".
testing naddy, me and others,
ok deraadt@
In the pledge_namei() callback, stop looking for the libc special files
when the open(2) system call was used, and only do this for __pledge_open(2)
which libc will use when it intends to access those files.
This breaks compatibility in a pretty major way, but that's the price being
paid to undo this historical shortcut I made to satisfy libc's internal
requirements and make pledge(2) a possibility.
All the other parts are already in the tree. There may be a few dangling
issues in ports or base which we'll find over time.
Add support for BCM575xx devices, variously known as Thor or P5.
There are a few significant differences to earlier devices.
The nic now requires some host memory to use as backing store for its queues,
and for now we're overallocating to some extent. It's not a noticeable amount
of memory for a system with one of these nics in it, so this isn't a huge
concern.
P5 devices have notification queues to act as an indirection between tx/rx
completion rings and msi-x vectors. We set up one per queue and statically
map them to msi-x vectors in turn according to the intrmap.
The doorbell structures are now 64 bits, and all written to through the same
memory address.
Ring groups are not used, so the functions to allocate and free ring groups
don't do anything for P5 devices; instead, rings are directly associated
with each other on creation, and aggregation rings are identified by a
different ring type.
[3 lines not shown]
Draw message as one format, allowing prompts and messages to occupy only
a portion of the status bar, overlaying the normal status content rather
than replacing the entire line. A new message-format option now controls
the entire message (like status-format). From Conor Taylor in GitHub
issue 4861.
