Shouldn't be needed, but install has no other good way to deal with
this.
Pointed out by Rob Windsor in PR 14394 -- I committed his patch plus
one for something he didn't hit yet.
struct ieee_double, rather than a pointer cast. This seems to enable
GCC 2.95.3 to get the instruction dependencies right (the old one fell
foul of ANSI aliasing rules), and it also generates more sensible code in
general.
If this is the correct solution, it should be applied to the other ports.
If it's not, someone should come up with one that _is_ correct.
- Remove old references to new toolchain system in favor of reference to
BUILDING.
- Remove old reference to MKCRYPTO being added; new .mk files are used
at the top level automatically.
- Clean up vertical space.
device drivers:
- Various native device entries in cdevsw/bdevsw.
- Rework the interrupt infrastructure to provide more flexibility to
the platform-dependent back-end. Rewrite the "ofwgen" simulated
interrupt routines to reflect the changes.
- Clear out the BAT registers and set the fixed battable entries before
calling the platform init routine. The platform init routine is allowed
to set entries in the battable.
- Don't call the platform cons_init routine until after translation is
enabled -- we might need translation to work in order to access bus
space.
- calculate the offset and length of the postbl before byteswapping.
problem noted by der Mouse.
- use offsetof() to determine # of fields to calculate in initial
loop, rather than hard-coding in `52 fields'
- improve comments.
check on the correct node
- apply_specentry(): if this node is a duplicate of another, apply the
changes to the `master' entry instead of this one.
- fix inotype() to DTRT
- comment out some debugging info that is too verbose
aren't going to do anything with the information anyway and there is
error checking later anyway.
We can now succeed in creating symlinks to locations that don't exist,
just as ln -s will let us do, and we can use install instead of ln -s
in several Makefiles. The code was written with the obvious intent to
let you do this but apparently it was never tested.
THAT accurate and microtime(9) is painlessly slow on i386 currently.
This speeds up small transfers much. The gain for large transfers
is less significant, but notable too.
Bottleneck was found by Andreas Persson (Re: kern/14246).
Performance improvement with PIII on 661 Mhz according to hbench (with
PIPE_MINDIRECT=8192):
buffersize before after
512 17 49
1024 33 110
2048 52 143
4096 77 163
8192 142 190
64K 577 662
128K 372 392