docs/comments: keep prose hardware- and vendor-neutral

Drop incidental cloud-vendor and instance-type names from committed docs and a
code comment; the technical content is unchanged (a generic-timer frequency
ceiling raised to 2GHz to cover current server ARM cores; debugging/perf notes
that describe the host generically).

Author: gburd

arch/aarch64/arm-clock.cc

@@ -43,8 +43,8 @@ class arm_clock : public clock {
 arm_clock::arm_clock() {
     asm volatile ("mrs %0, cntfrq_el0; isb; " : "=r"(freq_hz) :: "memory");
     /* spec documents a typical range of 1-50 MHZ, but real cores run higher:
-     * the ampere-1a runs at 1GHz and AWS Graviton3 (Neoverse-V1) passes a
-     * 1.05GHz generic-timer frequency through to the guest, so the previous
+     * some server ARM cores run at 1GHz and up (one Neoverse-V1 part passes a
+     * 1.05GHz generic-timer frequency through to the guest), so the previous
      * 1GHz ceiling aborted the boot.  Allow up to 2GHz to cover current server
      * ARM cores with headroom while still catching a garbage read. */
     if (freq_hz < 1 * MHZ || freq_hz > 2000 * MHZ) {

modules/open_zfs/FINDINGS-osv-openzfs.md

@@ -1,7 +1,7 @@
 <!-- Copyright (C) 2026 Greg Burd -->
 # OpenZFS-on-OSv: debugging findings
 
-Hands-on debugging on EC2 m5d.metal, fedora:39 build container, KVM guests.
+Debugging notes from bringing OpenZFS up on OSv under KVM.
 
 ## Bug 1 - zpool export/import lfmutex owner assertion (FIXED)
 

modules/open_zfs/PERF-osv-openzfs.md

@@ -15,7 +15,7 @@ Both built from `pr/openzfs-draft` (1a298e1b), same commit, separate clones.
 
 ## Test bed
 
-- Host: AWS `m5d.metal` (96 vCPU, 377 GiB RAM, local NVMe instance store).
+- Host: a 2-socket x86-64 bare-metal host (96 vCPU, ~377 GiB RAM) with local NVMe.
 - Guest: OSv `zfs_builder.elf` booted directly (`--nomount --noinit
   --preload-zfs-library /zfs-bench.so <wl>`), **8 GiB RAM, 4 vCPU, KVM**.
 - Backing: