Larry Cohen wrote:
>
> Just trying to get the final answer on trying to reduce the size of core
> dumps (my 250 megabytes dumps
> fill up a disk really fast).
> I think I read two conflicting messages about this. The FAQ says that
> the entire memory has to
> be dumped but in the archive I see a message from Dave Winchell that
> Mission Critical Linux's mcore
> can selectively dump pages which dramatically reduces the dump sizes.
> I also noticed that I could further reduce the LKCD dumps with gzip (by
> 50%). Would it be possible at least
> to improve the compression by using zlib ?
Sure -- I can add this as an option, but I'll have to see which
code I can use in the kernel as the system's coming down. Let me
look at this. I did at one point, as there was something in the kernel
that was using zlib, but it wasn't a guaranteed part of the kernel and
I didn't want to make the vmdump.o module any bigger than it already
was. I'll go back and take another look.
There's also a patch to add in selective pages. I have to go back into
the archives and add it.
I'll also look at the APIC code. I'm out on business right now, so I've
got nothing better to do in the evenings but code ... :) The last word
I got out of Andi is that smp_send_stop() was sufficient, but again,
2.2 vs. 2.4 is different.
My list of things to do are:
1) Add the selective page dump code
2) Look into ACPI problem
3) Add fixes for module code to libklib if Tom doesn't do
it first
4) Add CONFIG_DISCONTIGMEM code options
5) Add rest of alpha system dependent code
--Matt
> Its been a challenge but with the hardware I have I could not get mcore
> working.
> In order to get lkcd working I had to comment out code in
> arch/i386/kernel/apic.c.
>
> void disable_local_APIC(void)
> {
> unsigned long value;
>
> clear_local_APIC();
>
> /*
> * Disable APIC (implies clearing of registers
> * for 82489DX!).
> */
> #ifdef notdef
> value = apic_read(APIC_SPIV);
> value &= ~(1<<8);
> apic_write_around(APIC_SPIV, value);
> #endif
>
> I'm not really sure why this works or what the side effects are. But if
> I dont do it I the system will
> hang trying to write out the dump header.
>
> -Larry Cohen
|