bootstrapped by `make world'. The version just built in ".."
normally won't work if the target system is not binary compatible.
The bootstrapped version has a better chance of working.
This makes the fixes and bugs in the previous 3 commits irrelevant.
Rev.1.11 was just wrong and rev.1.10 became unnecessary when
perl/perl was added to build-tools. Don't expect to build perl/usub
without using `make world' or equivalent if you don't have perl
installed.
bootstrapped by `make world'. The version just built in ".."
normally won't work if the target system is not binary compatible.
Don't build or install anything if _BUILD_TOOLS is defined. Then
we only want to build and install the mklocale binary, but the layout
of the mklocale tree forces recursing to mklocale/data for at least
the obj target even when _BUILD_TOOLS is defined.
bootstrapped by `make world'. The version just built in ".."
normally won't work if the target system is not binary compatible.
Don't build or install anything if _BUILD_TOOLS is defined. Then
we only want to build and install the colldef binary, but the layout
of the colldef tree forces recursing to colldef/data for at least
the obj target even when _BUILD_TOOLS is defined.
especially on a new install, where /var/msgs/bounds doesn't exist. I
moved my bounds file out of the way to create this before and after
on a quick 'n' dirty hack, which is probably the 23rd best way to do it,
but it works:
PR: 6963
Submitted by: Matthew Fuller <fullermd@mortis.futuresouth.com>
working when the target system is not binary compatible. Use various
hacks to work around minor problems in the source and binary tree
layouts:
- caesar and strfile are built normally (the source layout is good),
then installed by copying them to ${WORLDTMP}/usr/bin (they are
installed in ${WORLDTMP}/usr/games, but I don't want to put that
in $PATH).
- colldef and mklocale are built and installed normally. Messy and
incomplete relative path searches for them and caesar and strfile
can now go away.
- internal tools that aren't installed are now built and left lying
around for the `make all' pass to use. If the target system is
not binary compatible, it is critical that these tools don't get
rebuilt. Cleaning of the obj tree before building the internal
tools should ensure this.
- most internal tools are built using internal build-tools targets,
but tn3270 is simpler for a change - it has all the tools in a
separate tree, so they can be built using `make all'.
`#if defined(ONE_THING)' is a style bug, and i386 instead of __i386__
is a bug, since i386 is never defined when the kernel is compiled
by with the default flags (`gcc -ansi ...'). Here the bug disabled
the call to pmap_setvidram(), so ISA video memory was not mapped
WC on 686's. The bug may have been masked by bugs in the committer's
version of gcc - `gcc -ansi' incorrectly defines i386 for gcc = the
version of egcs on the 2.2.6 cdrom.
bits. We used a private, wrong, version of `struct dirent' to help
break getdirentries(), and we use a silly check that the size of this
struct is a power of 2 to help break mount() if getdirentries() would
not work. This fix just changes the struct to match `struct dirent'
(except for the name length).
not actually work for cross compiling, but that is another problem.)
Honor LDFLAGS for building internal tools. (Tools should normally
be built static to avoid problems with picking up target shared
libraries. bsd.kmod doesn't set -static yet, and has some problems
with `LDFLAGS=-static ...' in the environment.)
Honor LDFLAGS for building internal tools.
Always build intermediate object files explicitly so that binaries don't
change when they are rebuilt.
Fixed some style bugs.
Restored a variant of explicit rule for `setup', with modifications to
always build setup.o so that `setup' doesn't change every time it is
rebuilt because it has a temporary file name in it.
Removed explicit dependencies of foo.o on foo.c. These were mainly
placeholders for comments about missing dependencies of tools objects
on headers. This problem needs to be handled more generally.
Fixed the type of the string table size variable to match its use
(assuming that int32_t is 4 bytes and other unportable things).
bfd uses `unsigned char string_chars[BYTES_IN_WORD]', where
BYTES_IN_WORD can be 4, 8, or perhaps even 2 or 3, but it is
assumed to be precisely 4 bytes here.
Fixed printf format errors (don't assume that n_value in struct
n_list has type u_long, since it should have size BYTES_IN_WORD
and longs may be longer than words).
them as ints. Among other bugs, doing so at best caused benign
overflow followed by fatal sign extension on machines with 32-bit
ints and 64-bit longs.
writes of size (100,208]+N*MCLBYTES.
The bug:
sosend() hands each mbuf off to the protocol output routine as soon as it
has copied it, in the hopes of increasing parallelism (see
http://www.kohala.com/~rstevens/vanj.88jul20.txt ). This works well for
TCP as long as the first mbuf handed off is at least the MSS. However,
when doing small writes (between MHLEN and MINCLSIZE), the transaction is
split into 2 small MBUF's and each is individually handed off to TCP.
TCP assumes that the first small mbuf is the whole transaction, so sends
a small packet. When the second small mbuf arrives, Nagle prevents TCP
from sending it so it must wait for a (potentially delayed) ACK. This
sends throughput down the toilet.
The workaround:
Set the "atomic" flag when we're doing small writes. The "atomic" flag
has two meanings:
1. Copy all of the data into a chain of mbufs before handing off to the
protocol.
2. Leave room for a datagram header in said mbuf chain.
TCP wants the first but doesn't want the second. However, the second
simply results in some memory wastage (but is why the workaround is a
hack and not a fix).
The real fix:
The real fix for this problem is to introduce something like a "requested
transfer size" variable in the socket->protocol interface. sosend()
would then accumulate an mbuf chain until it exceeded the "requested
transfer size". TCP could set it to the TCP MSS (note that the
current interface causes strange TCP behaviors when the MSS > MCLBYTES;
nobody notices because MCLBYTES > ethernet's MTU).
