Rebuild when RAMROOT is toggled, and add an opt-in KVM target

`make ramroot` wiped the objects on the way in, and nothing wiped on the
way out. So a plain `make` afterwards kept every RAMDISK_ROOT object and
linked the 24 MB ramdisk back into a kernel that no longer used it: 25 MB
where a clean build is 836 KB. It still booted -- from the ramdisk -- and
the only outward symptom was stage2 having thirty times as much to load
through PIO, one sector at a time. That is the "booting got slower"
report this came from.

A stamp file records which setting the objects were built with, and the
tree is dropped when it changes in either direction.

qemu-kvm is opt-in and the default stays TCG, because the suite's timings
were established there. Measured with the guest timing itself,
`ls /bin | wc -l` goes 2.9s -> 1.1s. Boot does not improve and gets
slightly worse: every PIO port access in stage2 is a VM exit under KVM.

Retracts a claim from earlier today. The "2.2x speedup" attributed to the
region-based allocator was a measurement artifact -- the probe sent a
command, slept a fixed 50s on the host, then read /proc/uptime, so it
timed the sleep rather than the command, through a guest clock that drifts
under load. Bracketing the command inside one guest command line shows no
measurable difference. The e820 change is still correct; it is simply not
a performance change.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

.claude/memory/schos-bootcheck-trust.md

@@ -4,7 +4,7 @@ description: bootcheck silently stopped running and five of seven modes asserted
 metadata: 
   node_type: memory
   originSessionId: 9d8a2aef-bcf0-47a0-96c6-f706efab4a95
-  modified: 2026-08-09T19:17:56.770Z
+  modified: 2026-08-09T20:44:14.569Z
 ---
 
 **A green run is only green if you counted the checks.** On 2026-08-09 the
@@ -83,5 +83,27 @@ confounds each produced a confident wrong answer in one session:
    `make build/schfs.img` before every timed boot.
 3. **Ordering / host drift** — alternate the variants and run each twice.
 
-Time it from inside: `cat /proc/uptime` either side of the command. The guest's
+Time it from inside — **and bracket the command in a single guest command line**:
-tick does not care what the host is doing. See [[schunix-verify-by-driving-shell]].
+
+```sh
+cat /proc/uptime; ls /bin | wc -l; cat /proc/uptime
+```
+
+Sending the command, sleeping a fixed interval on the host, then asking for
+uptime **measures the host's sleep, not the command**. That produced a confident
+and completely fictional "2.2x speedup" for the e820 work: the two readings were
+50 s apart because the host slept 50 s, and the arm that looked faster only had
+a guest clock lagging under load. One command line, two readings, no host timing
+in the loop.
+
+4. **The guest clock drifts under TCG.** Missed timer ticks mean `/proc/uptime`
+   undercounts during heavy load, so it is trustworthy for bracketing a short
+   command and not for comparing long intervals across runs.
+
+Baselines, measured correctly 2026-08-09 with an 836 KB kernel:
+`ls /bin | wc -l` **2.9 s TCG / 1.1 s KVM**; boot to `schinit: ready` ~3.7 s TCG.
+`make qemu-kvm` is the opt-in KVM target. **KVM does not speed up boot** and
+slightly slows it: stage2 loads the kernel by PIO one sector at a time, and each
+port access is a VM exit.
+
+See [[schunix-verify-by-driving-shell]].

.claude/memory/schos-uefi-loader.md

@@ -4,7 +4,7 @@ description: "boot/uefi works under OVMF; the SCH64 entry, the GDT trap, and the
 metadata: 
   node_type: memory
   originSessionId: d4821aa9-2e49-40da-91e7-90ef9ed9997e
-  modified: 2026-08-09T20:03:43.364Z
+  modified: 2026-08-09T20:44:02.893Z
 ---
 
 **`boot/uefi/` boots.** `make uefi`, `make esp`, `make qemu-uefi`. Verified
@@ -61,12 +61,14 @@ raising the bound past a hole.
 
 Result: 1018 MB in 7 regions under OVMF, 1023 MB in 2 under BIOS.
 
-**It is also a large speedup, which was not the goal and is worth knowing.**
+**A "2.2x speedup" was claimed for this change on 2026-08-09. It was wrong,
-Measured from inside the guest (`/proc/uptime`, fresh image each boot,
+and the retraction matters more than the claim did.** The measurement sent a
-alternating rounds): boot to prompt **14.8 s vs 21.2 s**, and a cold
+command, slept a fixed 50 s on the *host*, and then read `/proc/uptime` — so it
-`ls /bin | wc -l` **23.6 s vs 51.2 s** — 2.2x. Both rounds agreed within
+timed the sleep, not the command, through a guest clock that drifts under load.
-0.3 s. So if the allocator ever looks slow again, this is the baseline to
+Correctly measured (see [[schos-bootcheck-trust]]), `ls /bin | wc -l` is **2.9 s
-compare against.
+under TCG, 1.1 s under KVM**, and the region-based allocator has no measurable
+effect on it either way. The change is still right; it simply is not a
+performance change.
 
 ## UEFI reaches userspace — 2026-08-09
 

Makefile

@@ -320,6 +320,26 @@ else
 RAMDISK_OBJ =
 endif
 
+# Drop the objects when RAMROOT is toggled, in *either* direction.
+#
+# RAMROOT changes CFLAGS, and make cannot see that: an object compiled one way
+# is happily reused the other way. The `ramroot` target already wiped on the way
+# in, and nothing wiped on the way out -- so a plain `make` after `make ramroot`
+# kept every RAMDISK_ROOT object and linked the 24 MB ramdisk back in. The
+# result still booted, from the ramdisk, and the only outward symptom was that
+# stage2 now had 25 MB to load through PIO one sector at a time. Measured:
+# 25,413,120 bytes against 836,840 for the same kernel built clean.
+#
+# Evaluated when the Makefile is read, so the wipe happens before anything is
+# compiled against the wrong setting.
+RAMROOT_STATE := $(if $(RAMROOT),on,off)
+RAMROOT_STAMP := $(BUILD)/ramroot.state
+$(shell mkdir -p $(BUILD); \
+        if [ "$$(cat $(RAMROOT_STAMP) 2>/dev/null)" != "$(RAMROOT_STATE)" ]; then \
+            rm -rf $(BUILD)/kernel $(BUILD)/lib $(BUILD)/cmd; \
+            printf '%s' '$(RAMROOT_STATE)' > $(RAMROOT_STAMP); \
+        fi)
+
 # -----------------------------------------------------------------------
 # Top-level targets
 # -----------------------------------------------------------------------
@@ -856,6 +876,20 @@ $(BUILD)/fs.img: $(MKFS) $(BUILD)/sysroot.stamp
 # -----------------------------------------------------------------------
 # QEMU
 # -----------------------------------------------------------------------
+# The same thing with hardware virtualisation. Opt-in, never the default: the
+# test suite's timings were established under TCG and changing the execution
+# engine underneath them is a variable to introduce on purpose.
+#
+# Worth knowing what it does and does not buy. Measured with the guest timing
+# itself, `ls /bin | wc -l` goes from 2.9s to 1.1s. Boot does *not* improve --
+# it gets worse -- because stage2 loads the kernel by PIO one sector at a time,
+# and under KVM every one of those port accesses is a VM exit, where under TCG
+# it is cheap.
+qemu-kvm: schos.img $(BUILD)/schfs.img
+	$(QEMU) -enable-kvm -serial mon:stdio $(QEMU_DISPLAY) -cpu host \
+	    -hda schos.img -hdb $(BUILD)/schfs.img \
+	    -m 1G -smp 2 -machine pc-i440fx-9.2
+
 qemu: schos.img $(BUILD)/schfs.img
 	$(QEMU) -serial mon:stdio $(QEMU_DISPLAY) -cpu qemu64,+rdtscp \
 	    -hda schos.img -hdb $(BUILD)/schfs.img \
@@ -955,7 +989,7 @@ qrtest: $(QRTEST)
 qrcheck: $(QRTEST)
 	$(PYTHON) tools/python/qrcheck.py
 
-.PHONY: all kernel ramroot cmd ports sysroot-install qemu \
+.PHONY: all kernel ramroot cmd ports sysroot-install qemu qemu-kvm \
         schfs-test schfs-objtest schfs-image-test schfs-check mkschfs \
         schfs-fsck crashshot qemu-schfs iso qemu-iso clean distclean \
         logo qrtest qrcheck crashqr usb-ids