User Details
- User Since
- Mar 11 2014, 8:46 PM (662 w, 21 h)
Yesterday
Mon, Nov 16
To be clear, the whole stack is approved by me. I haven't done detailed review of the actual annotations themselves, but I'm fine if you merge them all with me as an "Approved by" assuming you've tested them (which you already describe doing both in comments and the commit logs). You might want to explicitly note you tested the amd64 ones with kgdb which might preempt some followup questions you might otherwise get.
In general I think this series is fine to merge, though my suggestions about the commit logs on the amd64 review I think are relevant for all the MD commit logs.
Some suggestions on the commit log, maybe something like:
I think I agree with markj@. There aren't that many drivers you'd need to update for this? Ideally you'd mark the buffers after the assembly routines return using buffer/length passed to the assembly routines.
Sat, Nov 14
Widen locking in nvmf_controller_error and remove overly-clever double-check
Folding in extra header trimming
Thu, Nov 12
Replaced by D61468
Update for libnvmf now owning sockets and various other refactors for later tests
Update for libnvmf now owning sockets and various other refactors for later tests
Update for libnvmf now owning sockets and various other refactors for later tests
Update for libnvmf now owning sockets
Hold lock across terminate task scheduling
Mon, Nov 9
After trimming some more bits from libkvm in userspace, I've done another round of trimming in sys/i386/include. For now you can find the "new" updates in a fixup commit here: https://github.com/freebsd/freebsd-src/commit/825e61d8154dd9913d2dc92fa9f74f7fa060dca2 The interesting changes perhaps are the removals from <machine/cpufunc.h> as potentially some of those could be used in userspace in ports maybe (but not likely).
I guess this is ok to do as a separate commit, but I think it makes more sense to fold this into the 3rd commit as it isn't meaningful until then. Also, I think in your last commit, it's probably a bug that you don't expand pos_irq to a full u_int since _PRS can also use extended IRQ types.
