BSD Commit Log Search

   Allow vmd(8) virtio hardware init to fail so we can clean up.

   Remove a bunch of fatal() paths so we can return a failure when trying to
   initialize virtio devices. This lets us clean up any child processes
   created when we fork/exec things like virtio block and network devices.

   This is part of my overall effort to make vmd(8) not just abort and instead
   clean up its messes before exit.

   ok mlarkin@

   Backout previous vmd(8) commit with no log message.

   *** empty log message ***

   make rport_start more like ifiq_input/if_vinput.

   this lets it call if_input_process, which in turn pushes packets
   back into the stack with the current softnet threads netstack.

   toplogy -> topology

   carry the netstack pointer through to if_vinput in vport_if_enqueue.

   this should encourage more route caching and tcp bundling.

   regen

   add MediaTek MT7925 device id

   regress/isakmpd/x509: bring this a bit closer to compiling

   There's a whole lot more to fix here, which I leave to someone else.
   At least this should get rid of breakage that I caused.

   isakmpd: do not reach into ASN1_STRING for the SAN

   ok beck millert

   revert changes to handle newer amd microcode

   mlarkin reports this leads to a crash in pmap_randomize_level/memcpy
   on EPYC 7773X

   drm/amdgpu: disable peer-to-peer access for DCC-enabled GC12 VRAM surfaces

   From Vitaly Prosyak
   d9db9abf6667a1ad2b827635123ecb63ea86184f in linux-6.12.y/6.12.59
   22a36e660d014925114feb09a2680bb3c2d1e279 in mainline linux

   drm/amdkfd: relax checks for over allocation of save area

   From Jonathan Kim
   325aa07165394b8e866ffe9ec1d4c99d9195b2f2 in linux-6.12.y/6.12.59
   d15deafab5d722afb9e2f83c5edcdef9d9d98bd1 in mainline linux

   drm/amdgpu: Fix NULL pointer dereference in VRAM logic for APU devices

   From Jesse Zhang
   43aa61c18a3a45042b098b7a1186ffb29364002c in linux-6.12.y/6.12.59
   883f309add55060233bf11c1ea6947140372920f in mainline linux

   drm/amdgpu: hide VRAM sysfs attributes on GPUs without VRAM

   From Christian Koenig
   39a1c8c860e32d775f29917939e87b6a7c08ebb1 in linux-6.12.y/6.12.59
   33cc891b56b93cad1a83263eaf2e417436f70c82 in mainline linux

   drm/amd/pm: Disable MCLK switching on SI at high pixel clocks

   From Timur Kristof
   d033e8cf4e8f6395102cdbc3cb00dc7cb9542f53 in linux-6.12.y/6.12.59
   5c05bcf6ae7732da1bd4dc1958d527b5f07f216a in mainline linux

   drm/amdgpu: remove two invalid BUG_ON()s

   From Christian Koenig
   eaf12bffd7f79f4d46ec028706f9d1a2d90f46fd in linux-6.12.y/6.12.59
   5d55ed19d4190d2c210ac05ac7a53f800a8c6fe5 in mainline linux

   drm/amd: Fix suspend failure with secure display TA

   From Mario Limonciello
   8c364a3a768f1abbfdd597f785a5c9c1407c7c72 in linux-6.12.y/6.12.59
   b09cb2996cdf50cd1ab4020e002c95d742c81313 in mainline linux

   drm/i915: Fix conversion between clock ticks and nanoseconds

   From Umesh Nerlige Ramappa
   d453865e6e1aa6e4f6af9371e84e1bdae8b05155 in linux-6.12.y/6.12.59
   7d44ad6b43d0be43d080180413a1b6c24cfbd266 in mainline linux

   drm/i915: Avoid lock inversion when pinning to GGTT on CHV/BXT+VTD

   From Janusz Krzysztofik
   4e73066e3323add260e46eb51f79383d87950281 in linux-6.12.y/6.12.59
   84bbe327a5cbb060f3321c9d9d4d53936fc1ef9b in mainline linux

   move mention of default MaxStartups (which uses the start:rate:full
   form) to after the full explanation of the start:rate:full form.

   GHPR568 from Santiago Vila

   Support writing ED25519 keys in PKCS8 format. GHPR570 from Josh Brobst

   avoid leak of fingerprint on error path; from Lidong Yan via GHPR611

   don't set the PerSourceNetBlockSize IPv6 mask if sscanf didn't
   decode it. From Mingjie Shen via GHPR598

   Intel now documents a minimum runtime update revision in the microcode
   header.  This is intended to help decide whether to do an update after
   userspace is already running.  As we only do updates early in boot just
   update the struct.

   https://www.intel.com/content/www/us/en/developer/articles/technical/software-security-guidance/secure-coding/minimum-runtime-ucode-update-revision.html

   give ssh-agent more time to start in tests; requested in GHPR602

   When testing PKCS11, explicitly allow the module path in ssh-agent.

   Allows testing of PKCS11 modules outside system directories.

   From Morgan Jones via GHPR602

   When loading FIDO2 resident keys, set the comment to the FIDO
   application string. This matches the behaviour of ssh-keygen -K

   From Arian van Putten via GHPR608

   move aggr/trunk input processing into ether_input.

   previously it replaced the ifp->if_input function pointer on ethernet
   interfaces, which is always ether_input, to intercept packets. this
   makes it work the same as bridges (bridge/veb/tpmr) for intercepting
   and filtering packets now.

   this makes it (more) possible to call ether_input without netlock
   (which serialised the modification/use of if_input), and makes it
   possible for ethernet interfaces with custom if_input functions
   that eventually call ether_input to work as trunk/aggr ports.

   Some AMD microcode now has a minimum version for updating with an MSR.
   AMD-SB-7033 claims a protection fault will be generated if not at that
   level.

   The microcode containers can now also have multiple microcode patches
   for a given family-model-stepping combination.  One with a minimum version,
   and one without.

   Handle multiple patches and check minimum version using part of the
   container header.  Limit this check to family 19h and above.  The header
   is not publically documented and this part seems to have another use on
   earlier CPU families.