BSD Commit Log Search

   More corrections...

   I was also looking at the build breakage fixed by the previous commit.
   I found another issue (non-build breaking) though:

   Free the results from prom_getprop() in the newly added code, if a
   subsequent malloc() fails (don't lose memory).

   Also properly indent a line whose preceding condition shifted right
   (existing code from previously, now not always executed).

   Feel free to redo these changes some other (non build-breaking) way.

   fix the build.

   fix various typos in comments.

   pidfile: set errno when reading a corrupt pidfile

   strtoi doesn't set errno itself but passed it back as an argument

   Bump timeout high enough to give slow machines with slow disks a chance
   to complete the tests.

   t_tcp_connect_port: these tests take a *long* time on systems with a
   single slow CPU, adjust timeouts.

   s/aceleration/acceleration/ in comments.

   Fix up PROM's where reg is encoded as a 64-bit and a 32-bit value
   (e.g. 00000400 0fc62020 00000010), but we want 2 x 64-bit values.
   U45 with OBP 4.21.2 has jbus-i2c configured like this.

   fix build.

   grrrrrrr rump

   pidfile: Fix a few issues with error paths.

   pidfile_unlock() returns -1 on error rather than EBADF which is now errno.
   pidfile_read() now closes the opened fd on error if not pidfile_fd.
   pidfile_lock() now closed pidfile_fd if failing to strdup to pidfile_path.

   A long standing issue correctly setting FD_CLOEXEC has been resolved in the
   case where O_CLOEXEC is not defined.

   doc: Changes for last 3 weeks.

   doc: Fix syntax (hopefully). Web parser doesn't seem to like spaces
   on continuation-line indentation.

   mi pmap: aarch64: ARMV81_HAFDBS support

   mi pmap: aarch64: add some more helper variables to pte_make_enter. NFCI.

   spaces to tab

   eqos: Allow RX descriptors to be reused earlier.

   While here, fix a typo.

   Restore if_schedule_deferred_start to keep transmit queue running.

   amigappc doesn't need its own (amiga's) copy of dkbad.c, either.

   Use correct name for bit 24 in dwc_gmac_desc_std_rx_init_flags()

   eqos: Replace TX kthread with softint.

   TX hangs were reported on RK3588 with the kthread and this performs a lot
   better for me anyway.

   eqos: Add IPv4/TCPv4/UDPv4 checksum offload support.

   libroken: Rename static pidfile_path to rk_pidfile_path

   This is odd because libroken uses pidfile(3) AND it's own pidfile system.
   So potentially it's managing two pidfiles.
   libutil.h owns the pidfile namespace (other OS expose other names which
   may conflict) so rename the local vars to rk_pidfile_*.

   G/C useless / unused references to <sys/dkbad.h>.

   pidfile: Add functionality for use in sandboxing

   pidfile_fd() returns the file descriptor of the locked pidfile so it can
   be limited in sandboxes like Capsicum.

   pidfile_unremoveable() will avoid unlink(2) being called at program exit
   to avoid tripping any sandboxes like Pledge which may restrict it or
   it's just not reachable anymore if the program is in a chroot(2).

   pidfile_path() returns the path of the locked pidfile so it can be removed
   by a process than can reach it if pidfile_unremoveable() has been called.

   pidfile_unlock() just releases the resources held by the locked pidfile.
   This is lighter weight than pidfile_clean() as it avoids a lseek and read.

   - Reduce delays for MII access.
   - Fix multicast filter setup (and thus IPv6).
   - Align MAC address setup code with other drivers.
   - Create bus_dma map with BUS_DMA_ALLOCNOW to prevent failures
     when loading the map.

   Add more register definitions.

   eqos: More performance improvements.

    - Move TX to a separate kthread and allow TX and RX to run under separate
      locks. Lazily clean up TX descriptors instead of requesting an interrupt
      for TX completions.
    - Use the hardware watchdog to implement RX interrupt coalescing.
    - Remove some redundant MMIO writes and bus_dma syncs.

   mi pmap: more debug

   KNF

   spaces -> tab