Re: Impression of 2.2.14 vs. 2.2.15-pre

[Rik van Riel]

On Mon, 24 Jan 2000, Steve VanDevender wrote:
> Paul writes:
>  > 	One other thing 2.2.15pre did was completely expose my bad memory:)
>  > Under 2.2.14, I had one suspicious compiler segfault in weeks up uptime
>  > and lots and lots of building. Right away on 2.2.15pre, I had a tarball
>  > untar with a character off by a bit, corrupt gcc tmp files, etc. I did
>  > some testing on my mem, and it is definitely flakey.... because of
>  > this, my testing was of limited duration...
> 
> I wish I could blame the problems in 2.2.15pre4 on bad memory, but
> I think it's pretty unlikely that my memory went bad at the same
> time I booted into it, and I haven't seen mysterious segfaults on
> this system for years.

There seem to be some problems with programs holding the
mmap_sem semaphore. People are working on it.

> I'm still pretty sure that the VM changes in 2.2.15pre4 have some
> kind of problem.  I'm back at 2.2.15pre3 and my system has been up
> for longer than it took to lock up under 2.2.14pre4.

The VM changes have _exposed_ problems. It turned out that
a lot of places in the kernel call __get_free_pages() with
GFP_WAIT set, even though they get into trouble if they
have to wait...

I've programmed something to catch the bad code and work
around it. I've been trying to break the kernel with this
code for a few hours now (_heavy_ stresstesting) and
things seem to be ok.

When it survives the last few tests I'll post the patch
and put it online at http://www.surriel.org/patches/.

regards,

Rik
--
The Internet is not a network of computers. It is a network
of people. That is its real strength.