Re: tmpfs replacement

[Roland McGrath]

> Thanks for your help. For completeness here is the "bt full" of thread
> 4, with debugging symbols for glibc this time, x/i and info regs.

That verifies what we suspected.  The libc symbols help a little, but there
is some bug in gdb or gcc that prevents the libpager/libdiskfs frames from
showing all the argument info, and that is still annoying.

> What I expect is that after the page is deattached from the memory the
> object is destroyed... That is what the test program I wrote made me
> think. Does that make sense or do you have another idea?

Sorry, this description doesn't really mean anything specific to me.
You'll have to be more precise.  There is a live mapping created just then
by the vm_map call in do_memcpy (pager-memcpy.c), that is what tmpfs's
thread is blocked on.  The failure of the memory access to either succeed
or produce an exception is the bug, and that must be in the pager/kernel.