Add sysctl flag CTLFLAG_TUN to loader tunables

[zlei]

Summary

These variables are actually loader tunables. Add sysctl
flag CTLFLAG_TUN to them so that sysctl -T will report them correctly.

No functional change intended.

MFC after: 3 days

[zlei]

Those are part of the tunables spotted by the following script:

#!/bin/sh

egrep -ohr 'TUNABLE_[A-Z]+_FETCH\(.+,' sys | egrep -o '".+"' | sed -n -Ee 's/^"(.+)"$/\1/p' | sort | uniq > tunables_uniq.txt

while read -r line; do
    sysctl -q $line > /dev/null && {
      t=$( sysctl -TN $line )
      [ -z "$t" ] && echo "$line"
    }
done < tunables_uniq.txt

Tested on an amd64 VM.

[kib]

sys/kern/kern_proc.c

165 ↗

(On Diff #128379)

What if enabling this tunable everywhere, might be in a follow-up commit?

[zlei]

sys/kern/kern_proc.c

165 ↗

(On Diff #128379)

It was loader tunable since edc8222303c78 (Make kstack_pages a tunable on arm, x86, and powepc.)

init_param1(void)
 {
 
+#if !defined(__mips__) && !defined(__arm64__) && !defined(__sparc64__)
+       TUNABLE_INT_FETCH("kern.kstack_pages", &kstack_pages);
+#endif

Then spark64 support was removed in 58aa35d42975c

init_param1(void)
 {
 
-#if !defined(__mips__) && !defined(__arm64__) && !defined(__sparc64__)
+#if !defined(__mips__) && !defined(__arm64__)
        TUNABLE_INT_FETCH("kern.kstack_pages", &kstack_pages);
 #endif

And finally mips support was removed in a3571129380d:

init_param1(void)
 {
 
-#if !defined(__mips__) && !defined(__arm64__)
+#if !defined(__arm64__)
        TUNABLE_INT_FETCH("kern.kstack_pages", &kstack_pages);

What if enabling this tunable everywhere, might be in a follow-up commit ?

It is good to enable it everywhere. That requires extra works to enable kstack_pages tunable on arm64, but I'm not expert on arm64 architecture and have no clues how to enable that feature.

[kib]

To enable this feature on arm64, you need to remove the remaining #ifdef braces. When I added the knob, I disabled it on arches where thread0 stack size cannot be controlled with it. I think it is not too important in current times, and arm64 maintainers can fix it eventually if they want.

[zlei]

Just leave a note here.
It seems riscv might have problem with kern.kstack_pages. The logic for thread0 stack size is identical to arm64.