I would suggest explaining the difference between -o cpu and -o core in the description of the -o option.

Sorry to nag, but did you get a chance to test this?

• Test.

• Test.

• Test.

• Test

• Disable casper in the rescue build.

• Get rid of the (inconsistently used) predicates and just use inline tests.

In D18342#393604, @landonf wrote:

This looks good to me; I'll try to test on real hardware over the weekend, just to be sure.

• Add xfopenat().

• Address feedback from cem.

In D17940#392523, @jchandra wrote:

markj: Regarding the question on pxm prefix. I used the convention starting with the filename. I kind of settled for acpi_pxm since I could not think of a better filename and prefix (which covers SRAT for resource affinity and SLIT for locality).

An alternative would be to have acpi_proximity.c containing:
int acpi_parse_proximity_tables(int cpus, vm_paddr_t maxmem);
int acpi_set_mem_locality(void);
int acpi_set_cpu_locality(void);
void acpi_free_proximity_data(void);

Let me know if you have any suggestion on this.

• Further cleanup based on feedback.

In D18464#393352, @jhb wrote:

This is inspired by PR 231515?

Update the zopen.3 man page.

Rebase.

Approved.

@pho has a number of gmirror stress tests. Peter, do you mind giving this patch a try?

Most of my inline comments are nits, but: I don't really understand how the wakeup() mechanism works with the per-CPU caches. Threads may remain asleep on "zonelimit" indefinitely if uz_items == uz_maxitems but active threads are able to avoid the slow path. Before, the ZFLAG_FULL check in the zfree fast path handled this. Is there anything preventing starvation?

• Fix an off-by-one.
• Retire INP_PCBLBGROUP_PORTHASH. It's identical to INP_PCBPORTHASH, and because we now size the hash tables accurately, it would be nice to consistently use the same port hash everywhere.

• Attempt to handle all of the bugs that Alan pointed out.

I suspect that the anon clustering will interact suboptimally with the jemalloc behaviour discussed in D16501 and elsewhere. In particular, jemalloc will unmap small regions of the address space, leaving holes. With clustering, those holes won't be reused since we no longer perform a first-fit search. IMO it would be worth reconsidering how anon_loc works; rather than advancing it after each successful clustering, maybe it should be constant after the initialization to a non-zero value, so that we attempt to fill holes with new mappings before extending the clustered region further. I do not think this needs to be done prior to commit though.

Approved

Address some of the feedback. I left replies for the rest.

Address remaining issues:

• Generalize the probe service to all unconnected sendmsg() calls. I had been creating one pre-connected socket per interface, but that only works if the interface is up when rtsold starts. If it comes up after we enter capability mode, we're stuck.
• We can't call getifaddrs() in capability mode. (The SIOCGIFINFO ioctl does effectively the same thing, though, and is permitted in capability mode.) For now, just add yet another casper service to get the address flags of the link-local address on a specified interface.

Explicitly bzero 32bit compat register structures in sys_ptrace().

• Zero out reg structs in procfs too.

This mostly LGTM, a couple of nits inline.