Address suggestions.

Bump PORTREVISION of modified ports.

In D47230#1093667, @carlavilla wrote:

Thank you so much for your changes.
I'm gonna make a review and if everything it's ok I'll make the commit :)

The change has already merged.

In D48805#1113409, @glebius wrote:

Maybe commit changes to ip_output.c and ip_fastfwd.c separately? Cause they are a generic change, not something specific to ICMP.

Split off network stack changes

In D45409#1113549, @dougm wrote:

The description of this patch includes:

With this change, the pool field of the first page of a free block is used by the buddy allocator to identify its pool, but the buddy allocator makes no guarantees about the pool field value for allocated pages.

So, the value of the pool field for allocated memory is undefined an irrelevant. Now, way back when I started this, I was advised to consider a future when there were more than two pools, and the pool fields of the first reservation page would tell me what pool to free into, and so some code that reflected that notion got written, and it wasn't purged effectively enough then the decision was made to just stick with the two pools and let some future person deal with a flood of new pools.

From my inexperienced eye, the problem looks like when we're transferring pages from another pool to the requested pool, each of the m_buddy in vm_phys_split_pages will have its pool updated, but nothing updates the pool of the page being allocated itself. vm_phys_set_pool previously did that, along with various other updates, but now that it's vm_phys_finish_init it does not.

The description of this patch includes:

In D47169#1113528, @diizzy wrote:

Actually this is incorrect, they NEED to be submitted on bugzilla for timeouts etc to apply

Change successful message

I reproduced the panic on main and submitted PR 284539

In D48796#1113489, @emaste wrote:

bsd.prog.mk has:

# Use ${DEBUGDIR} for base system debug files, else .debug subdirectory
.if defined(BINDIR) && (\
    ${BINDIR} == "/bin" ||\
    ${BINDIR:C%/libexec(/.*)?%/libexec%} == "/libexec" ||\
    ${BINDIR} == "/sbin" ||\
    ${BINDIR:C%/usr/(bin|bsdinstall|libexec|lpr|sendmail|sm.bin|sbin|tests)(/.*)?%/usr/bin%} == "/usr/bin" ||\
    ${BINDIR} == "/usr/lib" \
     )
DEBUGFILEDIR=   ${DEBUGDIR}${BINDIR}
.else
DEBUGFILEDIR?=  ${BINDIR}/.debug
.endif

presumably the affected test here is the first case that installs a test .so other than under /usr/tests/lib/....

Actually this is incorrect, they NEED to be submitted on bugzilla for timeouts etc to apply

I now see some panics on shutdown in CI, e.g. https://cirrus-ci.com/task/5507293774610432

Address @markj comments. Thanks, Mark!

Address @jrtc27 comments

Seems ok for me.

Add KASSERT()s, as suggested by glebius@.

OK. My preference would be to remove the date altogether, regardless of build settings, but this is an incremental step at least.

bsd.prog.mk has:

# Use ${DEBUGDIR} for base system debug files, else .debug subdirectory
.if defined(BINDIR) && (\
    ${BINDIR} == "/bin" ||\
    ${BINDIR:C%/libexec(/.*)?%/libexec%} == "/libexec" ||\
    ${BINDIR} == "/sbin" ||\
    ${BINDIR:C%/usr/(bin|bsdinstall|libexec|lpr|sendmail|sm.bin|sbin|tests)(/.*)?%/usr/bin%} == "/usr/bin" ||\
    ${BINDIR} == "/usr/lib" \
     )
DEBUGFILEDIR=   ${DEBUGDIR}${BINDIR}
.else
DEBUGFILEDIR?=  ${BINDIR}/.debug
.endif

I thought too much on this. @ae Are you OK with this revision ?

Think it twice, it is straight forward to synchronously detach the BPF descriptors. The logic is also simpler.

Thanks