Mark Lanteigne (ThirdIO) continued our research on #rowhammer \w #DDR4. 8 out of 12 modules vulnerable.
A must read. http://www.thirdio.com/rowhammer.pdf
-
-
@RichFelker as we said in our ccc talk... We don't prevent bitflips. We only need to make sure they can't occur across privilege boundaries -
@lavados is that distinction meaningful? part of the problem is HW can't detect victim rows w/ fine enough grain. or do you mean harden SW? -
@jawnv6 you can only hammer rows in physical proximity. OS memory allocators could prevent that and partially do that already today. -
@lavados but if you knew where the victim rows were, the MC could just trigger refresh. there's swizzling that prevents this total knowledge -
@jawnv6 there is not 1 victim row... on all dram i had access to i found thousands of possible victim rows.... -
@lavados just because you can empirically extract the info later doesn't mean it's available/usable to the relevant system agents at runtime -
@jawnv6 exactly my point. that's the exact reason why I wouldn't try to protect victim rows or refresh them more often or anything like that
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.