This is an archive of the discontinued LLVM Phabricator instance.

[mte] [nfc] do not keep Tag in AllocaInfo
ClosedPublic

Authored by fmayer on Feb 3 2022, 6:12 PM.

Details

Summary

this is to get the implementation closer to HWASan

Diff Detail

Event Timeline

fmayer created this revision.Feb 3 2022, 6:12 PM
fmayer requested review of this revision.Feb 3 2022, 6:12 PM
Herald added a project: Restricted Project. · View Herald TranscriptFeb 3 2022, 6:12 PM
fmayer updated this revision to Diff 405848.Feb 3 2022, 6:15 PM

keep old behaviour

fmayer updated this revision to Diff 405850.Feb 3 2022, 6:23 PM

rebase onto fix

fmayer updated this revision to Diff 405851.Feb 3 2022, 6:25 PM

saving files is hard

Not sure about this. I kind of like that isInterestingAlloca is called only once for each alloca - this way it can run heavier analysis (it does not right now) and it is also allowed to change its mind (ex. after we add some instrumentation to the function).

Not sure about this. I kind of like that isInterestingAlloca is called only once for each alloca - this way it can run heavier analysis (it does not right now) and it is also allowed to change its mind (ex. after we add some instrumentation to the function).

That's fair. I was torn between the two approaches (the other way would be to change HWASan to resemble the current state of MTE stack tagging). I thought this has less room for error from forgetting to check for interestingness (because it's not in the list to begin with), and because as you said isInterestingAlloca doesn't do anything particularly expensive.

eugenis accepted this revision.Feb 4 2022, 12:27 PM

LGTM

yeah nvm I'm ok with this

This revision is now accepted and ready to land.Feb 4 2022, 12:27 PM
This revision was landed with ongoing or failed builds.Feb 4 2022, 12:34 PM
This revision was automatically updated to reflect the committed changes.