From: obz@sisd.Kodak.COM (Orest Zborowski) Subject: (none) Date: Tue, 10 Mar 1992 23:03:17 GMT
i got 0.95 working last nite without much headache. i simply took the root
device, mounted it and moved all the files onto my hard disk. then i put the
kernel sources in another place and remade it, fixing the makefile along the
way. it compiled and came up without problem.
several notes:
o i'd like to be able to have a timeout waiting for the svga selection. in
the old system i never bothered to hit return to change the screen
resolution, so reboots were a single-step process. esp for machines which
want to run without operator intervention (i.e. with shoelace), they'd like
to automatically boot and come up with some default configuration.
o thanks, linus, for plugging in the fixes for file modes, although i'd still
argue that in open_namei() it shouldn't make the conclusion that
O_TRUNC ==> O_WRONLY|O_TRUNC. it seems it was built that way to support
the old mount command, which used only O_TRUNC in its open call.
o i made the (slight) mods to my mmap and io bitmap patches for 0.95. the
ega demo (which uses both) seems to compile and work fine. i will be posting
them anew (with all the pieces) as soon as some other things get resolved.
o bad news: i'm using the old gcc1.40 to compile stuff, and just like linus
predicted, i needed to use -fcombine-regs to compile the kernel. no prob and
the kernel seems to work ok. however, some of the x libraries now won't
compile, running out of swap space. i created a larger (6mb) swap area and
with 10mb of virtual mem i can't imagine running out of space, esp since
under the old system all of x compiled in 8mb of virtual space. tracking
it down it seems that the other processes' frames get evicted until the
compile's allocation reaches the maximum (in my case 640 frames). then it
goes to out of memory.
i saw in the sources some support for detection of thrashing (misspelled
trashing - a pun?:-) and was going to put in a test to see if the thrashing
test was causing some sort of failure. unfortunately i had performace a
make clean on my libs so had to wait forever...
o while compiling my linux-compatibility library i caused the compiler to
hang! this was a very small file which #included <math.h> then defined
cbrt(double d) { return pow(d, 1.0 / 3.0); }. the system ran but this
compile failed to stop. looking at the paging i saw that this process was
consuming a new page every 2 sec or so. this worried me because i was
running with the new io bitmap stuff, which adds entries into the task
struct. i throw this out to see if anyone has a similar problem. once this
gets resolved i'll feel more comfortable with the mmap and io bitmap mods
and will post them.
o mounting floppy filesystems no longer seems to work. when i mount, then
access it the kernel complains that the floppy has been changed and boots
the inode. oh well...
o i understand that 0.95 was compiled and tested using gcc2.0, and i'm in the
process of obtaining it, but it does bother me that the old compiler doesn't
work - things haven't changed that much that i can imagine would break a
compiler, esp if other old binaries seem to work fine.
have others had similar problems? sorry about being vague about some of these
things, it was late and i didn't have time to get more specifics. once i do
i'll post them.
zorst
obz@sisd.kodak.com