On 4/8/20 8:11 AM, Pavel Tatashin wrote:
On Wed, Apr 8, 2020 at 11:01 AM Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
From: "Matthew Wilcox (Oracle)" <willy@xxxxxxxxxxxxx>
I really don't like that this feature is called page poisoning.
We already had something called page poisoning and it's when you detect
a memory error in a page. This is just uninitialised pages. I don't
Hi Matthew,
Thank you for working on this. Uninitialized struct pages are often
zeroed by firmware, and there were a number of implicit assumptions
about that memory when I worked on deferred page initializations, this
is why it is important to also test when struct pages are specifically
set to a pattern that is not all zeroes, something that can happen
during kexec, or when memory allocated and freed by kernel during
boot. We have caught a good number of bugs using this mechanism. So,
this is poisoning, but I agree "page poisoning" name is misleading, as
we have this term used in another place. So, lets agree on a better
term: how about memmap poisoning (s/page_poisoning/memmap_poisoning/)?
Better to avoid the "poison" word entirely. It's just too well-established
as a hardware memory error case. Early ideas for other names, just to get
started: use "uninitialize" or "deinitialize" instead of "poison".
So approximately:
page_deinit
page_deinitialize
page_uninit
page_uninitialize
mem_deinit
...other variations...
thanks,
--
John Hubbard
NVIDIA
