o Fixed some minutes ago a critical bug in the asm of entry.S. This
was at least the cause of the UP stall. entry.S was using a
corrupted current-task-struct pointer.
o Optimized the scheduler. After a wakeup need_resched() must be
set unconditionally if the current process has pid == 0.
o Little little optimization in fork.c.
o schedule_timeout() updates. More smart in detecting misuse. Fixed
some place of the kernel that was eventually passing a negative
value to schedule_timeout(). It' s the caller that must check
that is passing a right value. The check inside schedule_timeout()
should be considered paranoid stuff to be _theorically_ removable
some day.
o Lastest lp driver. This new driver supports a new ioctl that
allow printer like HPs and Epsons to work _fast_ and with a very
little CPU usage. I write some tip about that in the top of the
driver sourcecode.
o Tons of time_*() jiffies updates.
o aic7xxx cleanup. Waste some byte of memory but it' s more clear.
Really it would be possible to save some more bit of memory
beeing equally clear without depends upon timer code lowlevel
implementation. I' ll eventually do that sometime.
o get_wchan() moved in the arch specific stuff. This will break
all no-i386 arch, but I think it' s the right thing and it' s
very useful to see the wchan info in the CTRL+bloc-scorr dump.
o My latest kswapd code. With this code kswapd is really a daemon an
runs in background and schedule as a normal process but with
dynamic priority. And more important my kswapd will not behave
differently on different harware because it doesn' t depend on
the jiffy/speed-of-machine relation.
o Swap cache garbage collector at exit time. It should make no
big difference and cause swap_free() to recurse one time but
since works fine and looks more precise I left it here.
o /proc/sys/net/ipv4/tcp_delayed_acks sysctl. Setting it to 0 you
can disable the delayed acks in the TCP stacks. In congestioned
networks envinroments this will cause the TCP stack to be more
responsive and eventually could decrease the congestion.
o mmap sync fix. This patch allow to share an mmapped region and
to update the shared mappings at sync/release time for all process
that are sharing such memory region. This improve things a lot.
It seems that for Tom Eastep this my feature was causing a
deadlock so I changed some bit that made sense and he now reported
me that arca-22 was just fine.
o proc/sysctl #ifdef fix. This will allow you to compile
your kernel without /proc and/or sysctl support. I' ve seen that
most of this stuff is been included in 2.1.129-prexx.
o little vmalloc fix.
o swap_cache_find_success/total now say something.
o fixed the show() routine in i386/kernel/irq.c (done by Ingo &
Linus).
o Fixed a subtle race in deleting the real time timer of every
process. (done and discovered by Linus)
o some other minor thing.
Everything in thist last patch worked fine so far here.
Note that when you' ll boot your new kernel you' ll run in the past for 1
hour (run uptime and you' ll see), so you' ll have exactly 1 hour to
stress test the kernel to eventually catch some nasty timer expire bug. I
had no one problem so far also while I was in the past though...
ftp://e-mind.com/pub/linux/kernel-patches/arca-23-against-2.1.128.diff.gz
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/