might be called without process context, leading to tsleep() panics
(This is a workaround. The timeout() driven wscons glue code
should go instead.)
Fixes tsleep() panics reported by Mark Brinicombe <mark@causality.com>
and Eric Haszlakiewicz <haszlaki@UAccess.NET>.
there are no pages available immediately to use as a PT page, don't
just roll over and die. Only do that if we're the kernel pmap (shouldn't
happen very ofter [ever?!], and I'm not certain we're guaranteed valid
thread context when operating on the kernel pmap). For user pmaps, wait
for the pagedaemon to wake us up when more free pages are available.
out-dated comments, and other unneeded stuff. This helps prepare
for cleaning up the rest of the code, and adding new functionality.
No functional changes to the kernel code in this commit.
Thanks to Scott for pointing this out to me (I got his mail and figured out
this change before seeing the discussion on tech-kern) and to Charles for
the initial explanation.
while a larger message buffer size might be a good idea (even as the
default), the existing definition here was way too large and was, in
fact, accidental!
- break anon related functions out of uvm_amap.c and put them in their own
file (uvm_anon.c). includes break up uvm_anon_init into an amap and an
an anon init function
- ensure that only functions within the amap module access amap structure
fields (add macros to amap api as needed)
- Don't use tcp_respond(), instead create the tcp/ip header from scratch,
and send it ourself.
- Reuse the mbuf that carried the SYN, or allocate one if that is not
available.
- Cache the route we look up to do the Path MTU Discovery check, and
transfer the reference to that route to the inpcb when the connection
completes.
* Macro'ize a small, but often repeated code fragment.
1. the mask is passed by value, not by reference
2. compare the rest of the mask bits not the valid ones
3. return 0, not the current fd's flags.
Now appletviewer works on an inet display. It will not work
with :0 because /tmp/.X11-pipe/X0 is a named pipe, not a
socket like our /tmp/.X11-unix/X0. Maybe I'll kludge svr4_sys_open
to return an fd to a unix socket when someone attempts to call
open on a unix socket.
- no longer conditionalized
- when traced, charge time to real parent, not debugger
- make it clear for future rototillers that p_estcpu should be moved
to the "copy" region of struct proc.
(Nonaka Kimihiro): grave accent and @ were the wrong way
-assign keysyms to the special japanese conversion keys (even if this
doesn't help wscons - but things must have names)
parents for children's p_estcpu. I think this problem has always been there.
It's particularly noticable with X because the server builds up non-trivial
CPU, and hence, non-trivial p_estcpu scheduler penalty. The repeatedly
forked children were always starting from scratch and receiving a scheduler
preference.
+ Don't use our own "clean" and "depend" targets
+ "make depend" works
+ Only include each include path once on compile lines
+ Clean up include file names
+ Don't build a separate libdrive.a, just specify driver source files
+ Use "make print-objdir" instead of old "printf ... | make -f-"
+ Remove more unnecessary targets, variables and other cruft
Still builds identical bootblocks to 1.3.3 with gcc, still too large
with egcs.
dec_boot.h, dec_exec.h & dec_prom.h were copied via respository
copy to sys/arch/pmax/stand/libsa.
version:
1. The stack_t returned before was SS_DISABLED; now we enable the stack
and return a 16K stack which is 16K under the current stack pointer.
This seems to be what solaris does.
2. Preserve uc_link.
of processes:
- Don't initialize rlim_max to RLIM_INFINITY. The limits for those should
be maxfiles and maxproc respectively. Programs expect getrlimit to
return reasonable values, so that they can allocate structures (for
example jdk does this).
- Don't initialize rlim_cur to NOFILE and MAXUPRC respectively, but to
min(NOFILE, maxfiles) and min(MAXUPRC, maxproc) respectively.
+ dependency on libsa works
+ set and use BINDIR/BINMODE
+ don't pull in sys/lib/lib{sa,kern}/Makefile.inc (``make clean''
doesn't make empty lib/{sa,kern} directories in each
directory/obj dir).
+ remove unnecessary targets, variables and other cruft
+ wrap lines at 80 chars
Re-ordering SRCS line in libsa/Makefile to a more logical sequence
(alphabetical by source dir) results in a libsa.a with archive members
in a different order, and a resultant bootrz that differs from the 1.3.3
bootrz. Untested as yet, so left in but commented out.
Also, _don't_ install the bootblocks (as per intent of previous commit)
or our own small libsa(!).
2.7.2.2-myc2 (and produces _identical_ bootblocks with the 1.3.3
bootblocks which weren't in the tree), but egcs 1.1.1 builds a ufs.o
that's about a 100 bytes too big, and it busts the 7.5k we've got for
bootrz.
Still needs cleanup - the Makefiles could be better/cleaner, and doesn't
install yet (because of above problem). At least there's something to
work with now.
Note that we have our own versions of a lot of the sys/lib/libsa
routines which are less functional/smaller.
bouyer