These should never get values large enough for sign to matter, but one
of them becoming negative could cause problems.
Details
Diff Detail
- Repository
- rS FreeBSD src repository - subversion
- Lint
Lint Passed - Unit
No Test Coverage - Build Status
Buildable 37922 Build 34811: arc lint + arc unit
Event Timeline
With a handy build complete, now I know of an extra instance. The
access in sys/arm64/arm64/gic_v3.c seems likely to be uneffected.
There actually aren't (yet) many places where this matters. One place it does matter is isrc_alloc_irq() which deals with the situation by having the "maxirqs" variable shadow intr_nirq. Since intr_nirq will never sensibly be negative it seems better for INTR_IRQ_INVALID to be a large positive value and testing whether an interrupt has a valid value be >= intr_nirq.
I note the isrc_irq entry of struct intr_irqsrc is unsigned. This actually effects D29310, since it means the test used needs a slight adjustment. This also allows D29326, which is getting rid of the shadow variable isrc_alloc_irq() has.
The 1 month anniversary of D29327. I think I've got appropriate people as reviewers, but no action...
I think this change is fine. I will tinderbox and commit it in a couple of days if no one objects.
With respect to D29310, I suspect it's not getting attention because it does indeed seems like a bikeshed. If the table is full, the problem will manifest as drivers not being able to attach, probably during boot, resulting in an unambiguous failure mode. Interrupt allocation isn't a dynamic operation, its performance is not important and doesn't warrant much scrutiny. Much more so for the unlikely case of interrupt allocation _after_ the table has been filled and some interrupts are freed.