sysctl_proc_args() did not check if uvm_io() returned no data, which
could result in uninitialised memory being returned. Rather than callers
of uvm_io() having to carefully check for this condition, treat it as an
error.
ok kettenis; initial report from Acts1631
Revert "drm/amdgpu: fix aperture mapping leak"
From Asad Kamal
a2e326c52c4bcecc033cd3ca2733fdbe30fbf55d in linux-6.18.y/6.18.45
b96c529cd2551b78316a4afa3237b2ed96ba03c8 in mainline linux
drm/amd/display: Check for tg ops in dce110_set_avmute
From Ray Wu
d089f32d34f821c8f0ef23d5fcd77bd43c1b3b92 in linux-6.18.y/6.18.45
3141e3d61469bba2624a91c5e2407f110b33b29e in mainline linux
drm/amd/display: Add AV mute wait frames to dce110_set_avmute
From Ray Wu
8aba384bfc8aac800335f0f458c4e77fc7223e48 in linux-6.18.y/6.18.45
443290d70b01e9c35830c300e3247c06581b594c in mainline linux
minor markup improvements:
in particular, more consistently use .Cm for module names, Makefile targets,
and fixed strings like Yes and No, .Ev for Makefile variables,
and .Sy for command names not documented in the present page
OK daniel@
SEV: Add option AMDSEV
To allow building the small RAMDISK kernel without SEV-* support
introduce the option AMDSEV.
When running a non-SEV kernel as a SEV guest we will raise a #VC
exception on the very first CPUID instruction. To allow a somehwat
graceful termination keep the locore0 #VC trap handler, fall through
to the termination code and issue TERMINATION request to the
hypervisor. In all other environments #VC will never be raised and
the kernel runs normally.
Build RAMDISK without SEV-* support, all other kernel keep SEV-*
support.
While there, cleanup and unify some definitions of the MSR protocol.
ok kettenis@
pf(4): connection counter for source track must use atomic_{inc,_dec}()
The connection counter is advanced when on state transition from
SYN_SENT to ESTABLISHED. That code in pf(4) currently runs with
no locks. If two packets happen to share the same source tracking
node, the race may occur. Two packets try to advance connection
counter simultaneously but only one operation is observed.
There is similar race between packet and timer that removes
expired state entry, the timer drops the connection count for
the source tracking entry while packet does opposite.
The result of those races may lead to connection counter underflow.
The issue was investigated and kindly reported by
Janak Trivedi <janakktrivedi _at_ gmail _dot_ com>
Feedback and suggestions by bluhm@
OK @bluhm
Update perl to 5.42.3
We already had the security changes, this is mostly version bumps
and documentation adjustments for them. There does pull in additional
changes in Archive-Tar, Compress-Raw-Bzip2, and IO-Compress.
While here, bump the libperl version for the previous CVE that
changed a public header.
OK and suggestions bluhm@
In mparse_open(), stop using the same local variable for two entirely
distinct purposes. Instead, declare two different variables, each with
an adequate type (const vs. non-const) and a more descriptive name.
No functional change.
Small style issue reported by gcc-16.1.1 on Fedora 44
via Xose Vazquez Perez <xose.vazquez at gmail.com>.
In mdoc_ptext(), stop confusingly declaring two pointers "const"
that actually point into a non-const string that is sometimes modified
in the function (to trim trailing whitespace).
No functional change.
Small style issue reported by gcc-16.1.1 on Fedora 44
via Xose Vazquez Perez <xose.vazquez at gmail.com>.
