On Fri, 26 Jun 2020, Alexandre Belloni wrote:
Hi,
On 25/06/2020 07:46:17+0100, Lee Jones wrote:
GENMASK and it's callees conduct checking to ensure the passed parameters are valid. One of those checks is for '< 0'. So if an unsigned value is passed, in an invalid comparison takes place.
Judging from the current code, it looks as though 'unsigned int' is the correct type to use, so simply cast these small values with no chance of being false negative to signed int for comparison/error checking purposes.
I've been thinking about that one but shouldn't the proper fix be in GENMASK? My understanding is that this happens because l is 0 and I don't think GENMASK would ever expect negative number. What about simply checking that h != l when l is 0?
Looks like Rikard Falkeborn recently submitted a patch to 'fix' this issue. However, Linus slammed it, decrying that it's the warning that's wrong, not the code. Looks like the type-limits warning might be getting downgraded to W=2 instead (although I don't see a patch yet).
I'm going to drop this patch.