BSD Commit Log Search

   use DL_PARTNUM2NAME()

   Use DL_PARTNUM2NAME() and DL_PARTNUM2NAME(). When we move to a-zA-Z
   partitions, the ascii letters are not linear so use partition numbers
   instead and translate the other way.
   ok krw

   use DL_PARTNAME2NUM(), and while here also check that diskmap isn't
   lying to us about an out-of-range partition

   ec_asn1_test: add an example using BLS12-377

   This exercises the cofactor guessing code with a large cofactor. Thanks to
   Daniel Bleichenbacher for pointing out this example. This contains a hack
   to use a bogus OID since this curve has none.

   use DL_PARTNUM2NAME and DL_PARTNAME2NUM instead of arithmetic relative to 'a'

   Use DL_PARTNAME2NUM and DL_PARTNUM2NAME instead of alphabet games

   use DL_PARTNAME2NUM and DL_PARTNAME2NUM instead of 'a'+ or -'a'.

   ixl/ice(4): use 128 segments for DMA maps of TSO packets

   This avoids unnecessary m_defrag() calls and gain some performance.

   with tweaks for kettenis and bluhm

   ok kettenis, bluhm

   remove unused wg_queue_len()
   ok rsadowski@ mpi@

   Emit the CCR BGPsec Router Key state hash in the JSON output

   OK tb@

   remove unused inskip()
   ok miod@

   Use DL_PARTNUM2NAME() instead of 'a'+ method

   rpiclock(4) and rpipwm(4)

   delete bogus maxpartition check

   Do not limit numeric host address conversion by family in resolv.conf

   Problem reported by yasuoka in sshd(8), which uses:
   AddressFamily any
   ListenAddress 0.0.0.0
   ListenAddress ::

   getaddrinfo(3) would only convert numeric host addresses for address
   families that are listed in /etc/resolv.conf and fail the others,
   leading to sshd(8) not listening on all address families.

   On the other hand, getaddrinfo(3) would convert all addresses if
   AI_NUMERICHOST was given, which is the correct behaviour.

   Initial diff provided by & ok yasuoka

   Temporarily support the O_DIOCGDINFO ioctl for 16-partition disklabels
   during the transition.  I think we only need it in these two drivers.

   Manually create an O_DIOCGDINFO ioctl value which supports the old
   16-partition disklabels as we go through the partition transition.
   This will allow an old fsck_ffs binary to operate in a new kernel.

   Introduce new DL_PARTNUM2NAME() and DL_PARTNAME2NUM() inline functions
   which will replace the unusable 'a'+partnum and partnum-'a' patterns
   throughout the tree.  These support a-zA-Z, but clamp to MAXPARTITIONS,
   and return -1 for out of range values.
   ok krw

   maxpartitions is currently 16, but we will be increasing it soon.  When that
   happens, the first device nodes for the "first drive" of a family (sd or wd or
   whatever) remain the same, but higher ones change in inconvenient ways, which
   affects the root partition.  This change creates /dev/root and /dev/rroot nodes
   which match the FUTURE dev_t layout, so that a user can fix their machine.
   We think this cross-over is going to be transparent for sysupgrade users, and
   users who build will have a good scheme for manual repair.
   ok krw kettenis florian

   Add rpipwm(4), a driver for the PWM controller on the RP1 chip.  Together
   with the previous RP1-related commits this makes the fan work.  Also based
   on eralier work by mglocker@

   ok on principle tb@
   ok mglocker@

   Add rpiclock(4), a driver for the clock controller on the RP1 chip.
   Based on an early draft bt mglocker@

   ok mglocker@

   Some cad(4) variants, including the one found on the RP1 chip, handle
   setting the correct clock frequency between 10baseT, 100baseTX and
   1000baseT in hardware.  Skip the calls to adjust the clock frequency on
   those.

   This diff is necessary because we're about to add a clock driver for the
   RP1 which means the calls to adjust the clock frequency will actually do
   something.

   ok mglocker@

   typo: tests -> test

   From Ivo van der Sangen

   fix uninitialized firmware path being used in qwx(4) error messages

   Always initialize amss_path in qwx_mhi_fw_load_handler(), not only
   if we are going to read the file from disk. The path will appear in
   error messages also in case loading a cached firmware image fails.

   qwx0: could not load firmware \M^@\M-Y\M-Q\^\\M^A\M-}\M^?\M^?\^Z':.\M-n\M-T\M^\\M-;P

   remove prototypes for removed functions

   bridge: use checksum offloading

   tested by bluhm

   ok bluhm

   Retire TCP flag TF_BLOCKOUTPUT.

   Together with socket splicing the flag TF_BLOCKOUTPUT was introduced
   in 2011.  It was needed to prevent direct calls from tcp_input()
   via sorwakeup() to tcp_output().  Due to unlocking TCP, socket
   wakeup always uses a splicing task on a separate thread, so
   tcp_output() runs deferred.  Also socket lock prevents that anything
   touches this TCP socket concurrently.  TF_BLOCKOUTPUT is not needed
   anymore.  This change also prevents some dangling TCP sockets in
   CLOSING state after running regress/sys/netinet/tcpthread.

   OK florian@

   wycheproof: provide PBKDF2 test harness

   Skip the tests for now since they increase the test's runtime by ~50%.
   A later commit will gate these tests behind REGRESS_SKIP_SLOW.

   regress parts for acme-client profile support, from florian

   vmd(8): fix a wrong comment

   comment had the wrong function name