https://chromium-review.googlesource.com/#/c/359631/

further cripples a relatively useless test. Since firmware is generally responsible for memory detection (and advertising that to the linux kernel), initially assigning to firmware test team.

Dmitry's suggestion sounds reasonable to me:
> I wonder if we should have a dictionary of "exceptional" boards, or somehow detect "full" chrome devices vs appliances.

I am proposing use of hwid (e.g. "GALE A2A-A2A-A2A-A2Q-A9G") to index into a table so get_mem_total() knows exactly what memory this hwid should have. Any given HWID should have unique mem size associated. Changing the installed RAM usually involves soldering chips. (Exception: AFIAK,only chromeboxes use SO-DIMMs but some older chromebooks might too - solution might be to return a list of acceptable RAM sizes).

If no entry exists, fail the test, then manually (and eventually automatically) create an entry. If one exists, fail if the reported physical memory ranges don't match.

While hand curation of the initial list is required, it's the only way I can see to enforce consistent and accurate discovery of physical memory.