arca-102 released

Andrea Arcangeli (andrea@e-mind.com)
Sat, 16 Jan 1999 01:20:11 +0100 (CET)


Name: arca
Version: 102
Kernelver: 2.2.0-pre7
Status: unknown
Author: Andrea Arcangeli <andrea@e-mind.com>
Description: Andrea Arcangeli's kernel tree
Date: 16-JAN-1998
Descfile-URL: ftp://e-mind.com/pub/linux/arca-tree/README
Download-URL: ftp://e-mind.com/pub/linux/arca-tree/

Differences with `Kernelver':

o New still a bit ;) experimental PG_dirty code, now it's shrink_mmap
that really swapout things, and the old swap_out() only generate
freeable/swappable pages, take care of the pte_young case and
swapout the shared mappings (vma->ops->swapout exists). Try it
and let me know!

o My new update_shared_mappings() code. This will greatly improve
performance while handling shared mmaped files with threading.

o My alternative cleaner IRQ_INPROGRESS handling. IRQ_INPROGRESS tell
us when really an irq is running with this patch applyed.

o irq_state array is PII cache aligned (32byte to decrease bus traffic
under intense irq activities in SMP)

o fixed xtime SMP race that was causing do_gettimeofday() to return
wrong retval

o getname buffer cached in the slab

o final fix for the console race I fixed some weeks ago. I developed
it some day after the first fix (the one that made atomic_t
atomic_bh_count). The day after I seen the bh_mask race too ;).
So we need a bh_lock (spinlock) to serialize disable/enable_bh().
This patch fix also a possible not critical race in remove_bh(),
btw note the wmb() to take care of SMP ;).

(o fixed a really very subtle console race that was causing a kernel
lockup if a proggy was writting to the console at the same time the
keyboard was pressed) just merged in 2.2.0-pre patches

(o fixed a bug in gettimeofday that was causing not monothonic
increasing values, improved also some hot things) merged in 2.1.132-3

o new IEEE1284 parport code from Tim Waugh.
It works fine here. It simplify radically the code of lp.c, the only
not very cool thing is that it decrease the way the printer can be
tuned via tunelp since IEEE1284 has its timings and its fixed way
to work... It's very fun though, so try it an be happy ;)

o get_wchan() moved in arch specific section and CTRL+SCROLL-LOCK
use get_wchan() to get the PC filed information. Due this change
the arca-56 kernel compile only on i386 and alpha (alpha is just
uptodate too).

o fixed page bound limits and the case of a new forked not
scheduled yet process in get_wchan()

(o fixed a little bug in a emergency SMP code in irq.c (Ingo and
Linus)) merged in 2.1.132-3

o some still pending (not yet merged from Alan) jiffies wrap update

o reimplemented hopefully completly wrap compliant time
to jiffies/msec (and reverse) functions

o removed unused flag from the mmstruct since we don't do aging
anymore (we use the PG_referenced bitflag only)

o new delayed ack heuristic invented by me to handle better
performance on congestioned networks, works fineee

o little vmalloc fix + vmalloc SMP thread safety

o bttv doesn't generate 50 irq per sec even if not used

o some minor things

Patch merged from other people:

o removed the map_nr field. This patch is from Max <max@Linuz.sns.it>.
It saves some bits of memory.

Andrea Arcangeli

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/