Re: [PATCH v3 00/40] mm: make VMA flag semantics explicit, eliminate VM_SPECIAL

Re: [PATCH v3 00/40] mm: make VMA flag semantics explicit, eliminate VM_SPECIAL

[Lorenzo Stoakes (ARM)]

< Edit mailing list, +sashiko as I gather this kind of feedback is wanted? :) >

I usually don't report it, but here's an example of some false positives, which
sashiko regenerates on respin.

Specifically the 2/40 (high) report was valid, as raised by Gregory and now
addressed ready for v4.

The ones on 1/40 (high), the other 2 on 4/40 (medium, medium) and the one on
24/40 (low) were all just wrong.

The one on 8/40 is bit but had some semblence of logic, but still not something
I'll take action on.

So a false positive rate of 83.3% here, or 67.7% if you exclude the 'meh' one.

In general I've continued to find the rate to be 50% at best, and it has not
improved over time.

It also seems that low/medium reports (which I think were disabled before?) are
particularly bad.

(I thought those were previously disabled for being inaccurate?)

I realise everybody is busy, but it'd be nice to have a means of marking
individual reports as hallucinations.

Thanks!

--
Cheers, Lorenzo