through an in-rumpkernel hypermemory allocator which knows it should
kick the pagedaemon and block in case ``waitok'' memory allocation
fails.
This allows us to recover from some out-of-memory situations.
Realworld'istically speaking (as opposed to whatever "should be"
theory), these OOM situations will happen extremely rarely if ever
when our hypervisor is a regular process. Speculatively, this
should be useful for other types of hosts.
issues remaining:
* the hypervisor does not know how to reclaim kernel memory (and
for the reason I stated above, I'm not sure if it makes sense
to teach the current implementation about that)
* vfs memory (buffers, vm object pages etc.) is not reclaimed
to find and unlink routes that reference the detached ifnet: make
if_rt_walktree() return ERESTART whenever it has deleted a route.
Whenever rt_walktree() returns ERESTART, if_detach() restarts it.
I believe that this fix resembles one by Jonathan Kollasch or by someone
else, which has languished in a PR for too long. Sorry!
Tested by me and by Jeff Rizzo.
XXX It's supposed to be safe for rn_walktree() to apply to the routing
XXX table a routine that may delete routes. Why isn't it safe in
XXX practice?
- Create and use symbolic constants.
- Convert some switch statements to functionally-similar caculations.
- Correct scheduling interval of high-speed isochronous transactions.
Previous calculation produced half the intended rate.
man pages to use mandoc unconditional as it gives reasonable output for
all man pages, not only a subset of mdoc(7). Use the newly installed
style.css for formatting and produce hyperlinks for .Xr.
These annotations help to mitigate false sharing on multiprocessor
systems.
Variables annotated with __cacheline_aligned are placed into the
.data.cacheline_aligned section in the kernel. Each item in this
section is aligned on a cachline boundary - this avoids false
sharing. Highly contended global locks are a good candidate for
__cacheline_aligned annotation.
Variables annotated with __read_mostly are packed together tightly
into a .data.read_mostly section in the kernel. The idea here is that
we can pack infrequently modified data items into a cacheline and
avoid having to purge the cache, which would happen if read mostly
data and write mostly data shared a cachline. Initialisation variables
are a prime candiate for __read_mostly annotations.
