Remove assumption that known type UIDs have a local alias
Currently this changes nothing, as all known types have a known
name (aka description), which is unlikely to ever change, and have
been assigned an alias for use with gpt -- which might change in the
future if we learn about types which no-one is ever likely to want
to use gpt(8) to create (but which might exist in tables processed).
Avoiding the assumptions is cheap (for both alias and name), so just
do it, and no-one will ever need to care in the future.
arm/imx23_olinuxino: implement multisegment DMA
This solves occasional issues in the sdmmc driver when it can't allocate large enough continguous segments due to memory fragmentation.
nvmm(4): Be extra careful with overflow checks so they don't get
optimized away by the compiler.
taken from DragonFly, Matthew Dillon, commit id
60ec4ed4a90ecd845abb48255a4cb4e904ebec49
ftp: simplify HTTP headers to use StringList
Implement custom HTTP header handling using the existing StringList,
instead of the less-portable SLIST_ macros from <sys/queue.h>.
Code is simpler too.
Emit the headers in user-provided order instead of last in first out.
mkimage: Write one block of padding up to 4MB.
Previously we wrote padding one byte at a time, which takes a while
when it's nearly 4MB of padding!
I believe these dd(1) options are POSIX-compliant. If the read from
/dev/zero is truncated, then conv=sync means it will be padded with
NUL bytes rather than reading more -- but that's fine; that's what we
would get by reading from /dev/zero anyway!
(Why did I make this round to a multiple of 4MB and _then_ add 1MB,
rather than the other way around of adding 1MB and then rounding to a
multiple of 4MB?)
No correctness change intended -- but this should improve performance
of the image build a little bit!
