Sorry it took me so long (washing dishes after christmas dinner takes much
longer thank breaking Linux ramdisk :) - I just reproduced it properly so
here are the steps:
1. compile 2.3.35-pre3 UP, 586TSC, ramdfisk as module with latest patch
posted by Frank.
2. insmod rd rd_size=25000 rd_blocksize=512
3. take a file stand.img, size=22M
4. write a shell script:
dd if=stand.img of=/dev/ram0 &
.....
dd if=stand.img of=/dev/ram5 &
instant oom. Now do the same with my version (using vmalloc()) -
everything is fine - all the excessive dd's just fail nicely with ENOMEM.
I think this is similar to a philosophical issue that goes back to
Linux+AIX vs SVR4+others wrt to VM "preallocation" policy, i.e. I think it
is better to use vmalloc() and refuse creating such large ramdisk in the
beginning instead of trying to protect too many buffers and end up with
OOM.
However, I say the above because I don't know of a clever way to restrict
using up too many buffers - hence I printed out 20,000 lines of code
yesterday in hope to read it all during xmas break :)
Regards,
Tigran.
On Sun, 26 Dec 1999, Linus Torvalds wrote:
>
>
> On Sun, 26 Dec 1999, Tigran Aivazian wrote:
> >
> > Fortunately thekernel is still
> > on 2.3.35-pre3 so I might be able to make the testcase ready before
> > Linus makes the next version (assuming he is asleep now).
>
> Hah! Sleep? I've heard of it.
>
> Linus
>
-
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/