From: Jeff Dike <jdike@ka...> - 2002-04-29 01:48:58
> so does this mean one should never allow the mem=X line to involve an
> X larger than your available /tmp space?
You can, but you can't use the excess. What'll happen is that UML will
poke at those pages to see if tmpfs is backing them. If you're over the
limit, it will get a SIGBUS, and those pages will be put off-limits for
> if you use tmpfs, it'll have as much space as you have free RAM+swap.
Not by default. There is a hard limit, and its default value is .5 RAM
From: Jeff Dike <jdike@ka...> - 2002-04-29 02:34:15
> I can easily put this to 96 or 64 and not affect the uml, would this
> result in free /tmp space for other programs on the host to use???
In the limit, it would. That would limit UML to using 96M or 64M of it,
leaving at least 32M or 64M for other things. However, if the UML is never
going to reach 64M/96M even when it has 128M available, it won't make any
From: Jeff Dike <jdike@ka...> - 2002-04-30 01:02:34
> That sounds like a bug? I wrote a quick test program to verify it and
> I agree that the kernel does different things for mapping /tmp/xzy
> than for /dev/zero.
Well, the current behavior does cause /dev/zero to live up to its name...