Add arm64(AArch64) support

[dzianisbely]

Add arm64 (AArch64) support

Summary

Adds arm64(armv8) support for the modern (5.9+) ftrace-based build. dattobd's x86-dependence turned out to be narrow. Changes isolated behind arch/capability conditionals, and leaves the x86 path untouched.

Supported kernels

arm64(AArch64): kernel 5.10+

ftrace_regs_set_instruction_pointer() exists only from 6.2 (renamed from ftrace_instruction_pointer_set; 6.2 is also where arm64 switched ftrace to WITH_ARGS). Hardcoding it would fail to build on ≤ 6.1 (implicit declaration). Detecting it and falling back to the WITH_REGS ->pc path lowers the arm64 floor to ~5.10.

Runtime-verified: 5.10 (Debian 11), 5.15 (Ubuntu 22.04), 6.8 (emulated), 6.17 (real HW). 5.9 is plausible (submit_bio path is 5.9+, ftrace_regs is 5.11+) but not tested.

P.S For building on 6.17+ kernels, I used ccflags-y := -g -Werror -I$(src)/../.. in /src/configure-tests/feature-tests/Makefile
Does not attempt legacy make_request_fn / syscall-table / armv7.

What changed and why

1. CR0 write-protect toggling is x86-only and is dead code on 5.9+ (called from nowhere). Wrapped the x86 body in #ifdef CONFIG_X86. no-op stubs otherwise. Closes X86 specific code failes building on ARM #248 (arm64 build break).
2. mount/umount hooks:
   • IP redirect moved to the dattobd_ftrace_set_ip() macro (the 3 cases above), gated on the HAVE_FTRACE_REGS_SET_INSTRUCTION_POINTER feature-test plus arch — not on the arch alone.
   • FTRACE_OPS_FL_SAVE_REGS requested only when CONFIG_DYNAMIC_FTRACE_WITH_REGS is present (x86 and arm64 ≤ 6.1 have it. arm64 6.2+ has only WITH_ARGS, where requesting SAVE_REGS makes register_ftrace_function() fail) — gated on the capability, not arch.
   • compile-probe configure-tests/feature-tests/ftrace_regs_set_instruction_pointer.c.
3. FENTRY_CALL_INSTR_BYTES value added. Its only for dattobd_submit_bio_real has no callers (dead code), so the value only needs to compile.
4. A REQ_OP_WRITE_ZEROES bio has no data pages. Tracer special-cases this, but the guard used the wrong macro, so on kernels ≥ 6.0 it compiled out. The page-less bio hit the per-segment loop, called page_get_inode() on a NULL page, and crashed kernel.
   Fix: key the guards off HAVE_ENUM_REQ_OP_WRITE_ZEROES, which checks the actual constant used and is true on 4.10+ for any arch. No change on < 6.0.
   Note: not arm64-specific, affects every 6.x kernel on all arches, including x86_64. Found on the arm64 port (ext4 lazyinit), reproduced on x86_64/7.0 with scsi_debug.

Evidence

Environments:

===== ENVIRONMENT (arm64, real 5.10 distro) =====
Sun Jun 21 17:39:32 UTC 2026
Linux arm64test 5.10.0-44-cloud-arm64 #1 SMP Debian 5.10.257-1 (2026-05-27) aarch64 GNU/Linux
PAGE_SIZE=4096
Debian GNU/Linux 11 (bullseye)
gcc (Debian 10.2.1-6) 10.2.1 20210110
/tmp/dall.sh: line 13: modinfo: command not found
--- feature-test (expect absent on 5.10 -> arm64 ->pc) ---
HAVE_FTRACE_REGS_SET_INSTRUCTION_POINTER not defined => arm64 ->pc fallback active

===== ENVIRONMENT (arm64, real 5.15 distro) =====
Sun Jun 21 17:04:59 UTC 2026
Linux arm64test 5.15.0-181-generic #191-Ubuntu SMP Fri May 22 19:27:05 UTC 2026 aarch64 aarch64 aarch64 GNU/Linux
PAGE_SIZE=4096
Ubuntu 22.04.5 LTS
gcc (Ubuntu 11.4.0-1ubuntu1~22.04.3) 11.4.0
version:        0.12.2
srcversion:     6AD47597211153C5E1728FB
vermagic:       5.15.0-181-generic SMP mod_unload modversions aarch64
--- feature-test result (expect absent on 5.15 -> arm64 ->pc branch) ---
HAVE_FTRACE_REGS_SET_INSTRUCTION_POINTER not defined => arm64 ->pc fallback active

===== ENVIRONMENT (arm64, real hardware) =====
Fri Jun 19 22:13:28 UTC 2026
Linux dbelyarm-QEMU-Virtual-Machine 6.17.0-35-generic #35~24.04.1-Ubuntu SMP PREEMPT_DYNAMIC Tue May 26 19:14:16 UTC 2 aarch64 aarch64 aarch64 GNU/Linux
PAGE_SIZE=4096
gcc (Ubuntu 13.3.0-6ubuntu2~24.04.1) 13.3.0
version:        0.12.2
srcversion:     841B21F4FF35C73387A0256
vermagic:       6.17.0-35-generic SMP preempt mod_unload modversions aarch64
#define HAVE_BDEV_FILE_OPEN_BY_PATH
#define HAVE_ENUM_REQ_OP_WRITE_ZEROES

===== ENVIRONMENT (arm64, emulated, 64K pages) =====
Sat Jun 20 08:41:11 UTC 2026
Linux arm64test 6.8.0-124-generic-64k #124-Ubuntu SMP PREEMPT_DYNAMIC Tue May 26 13:54:55 UTC 2026 aarch64 aarch64 aarch64 GNU/Linux
PAGE_SIZE=65536
gcc (Ubuntu 13.3.0-6ubuntu2~24.04.1) 13.3.0
version:        0.12.2
srcversion:     FD08148BF1FD432415BF50E
vermagic:       6.8.0-124-generic SMP preempt mod_unload modversions aarch64
--- module after build ---
srcversion:     A58570CFA61090438956EFF
vermagic:       6.8.0-124-generic-64k SMP preempt mod_unload modversions aarch64
version:        0.12.2
srcversion:     3A57035502391B4076B7EC9
vermagic:       6.8.0-124-generic-64k SMP preempt mod_unload modversions aarch64
#define HAVE_ENUM_REQ_OP_WRITE_ZEROES

---

• Build: clean, 0 source warnings, vermagic aarch64 on 5.10 / 5.15 / 6.8 / 6.17.
  build-arm64-5-10-4k.txt
  build-arm64-5-15-4k.txt
  build-arm64-6-17-4k.txt
  build-arm64-6-8-64k.txt.txt

• Full tests/ suite: 6.17 24 OK; 5.15 24 OK; 5.10 23 OK
  full-tests-arm64-5-10-4k.txt
  full-tests-arm64-5-15-4k.txt
  full-tests-arm64-6-17-4k.txt
  full-tests-arm64-6-8-64k.txt

• Runtime hooks: insmod OK, path_mount/path_umount ftrace hooks register, and a real mount fires handle_bdev_mount_event - i.e., the ->pc IPMODIFY redirect works.
  runtime-hooks-arm64-5-10-4k.txt
  runtime-hooks-arm64-5-15-4k.txt
  runtime-hooks-arm64-6-17-4k.txt
  runtime-hooks-arm64-6-8-64k.txt

• Explicit snapshot demo: a file created before the snapshot, then deleted from the origin, is still present in /dev/datto0 (content + 8 MB sha256 match) - COW correct.
  snapshot-demo-arm64-5-10-4k.txt
  snapshot-demo-arm64-5-15-4k.txt
  snapshot-demo-arm64-6-17-4k.txt
  snapshot-demo-arm64-6-8-64k.txt

• Real aarch64 HW (6.17, 6.8): setup-snapshot of the live root → /dev/datto0 (state 3). Full snapshot → image pull → restore-onto-live-root → clean reboot verified.