On 14/03/2026 16:36, Samuel Thibault wrote:
Mike Kelly, le lun. 09 mars 2026 22:04:19 +0000, a ecrit:
I've read parts of the paper written regarding porting Hurd to L4 and the
proposal to split user memory into guaranteed and pageable portions.
Actually we already have this supported, through the wired pages.
But what kind of memory does ext2fs need to allocate to release more
memory? Ideally the writeback path would not need additional memory at
all, or have pre-reserved memory that it can use. If that memory gets
swapped-out it's a concern of course, we could make root-owned ext2fs
wire it to avoid the issue.
I'll make a more detailed analysis when I can to be more specific about
what it's requiring memory for. That will help guide the way forward.
Might it make the resident set for rumpdisk smaller too if parts of that code
don't actually get used by Hurd?
I guess the problem with rumpdisk is rather that we have not actually
looked at its memory pool allocations, and have left it at very large
values which don't make sense for a userland process.
It has a resident set of over 200M on my machine and all (I think)
wired. I was imagining that some of it might not need to be wired or,
like you say, not needed at all.
Mike.
