CXL 디바이스 Core Dump 분석 — Device State·Mailbox Log·NUMA 토폴로지

1
# drgn 세션
2
$ drgn --core /var/crash/vmcore --vmlinux /usr/lib/debug/.../vmlinux
3

4
>>> # 모든 CXL port 나열 (개념적 — 자체 walker 또는 향후 추가될 helper)
5
>>> ports = walk_cxl_ports(prog)  # 직접 작성 or future helper
6
>>> print(f"{len(list(ports))} CXL ports")
7
3 CXL ports
8

9
>>> # Crash 시점의 region 상태
10
>>> for region in prog["cxl_regions"]:
11
...     print(f"region {region.name}: ", end="")
12
...     print(f"size={region.size:#x} ", end="")
13
...     print(f"state={region.state}")
14
region region0: size=0x4000000000 state=COMMIT
15
region region1: size=0x2000000000 state=COMMIT
16
region region2: size=0x0          state=ERROR     ← 의심
17

18
>>> # 의심 region의 decoder 상태
19
>>> r2 = prog["cxl_region_lookup"]("region2")
20
>>> for dec in r2.decoders:
21
...     print(f"  decoder{dec.id}: hpa={hex(dec.hpa_range.start)}")
22
...     print(f"    flags={dec.flags:#x}")
23
  decoder3.0: hpa=0x80000000000
24
    flags=0x4   ← ERROR flag set

1
>>> mbox = prog["cxl_memdev"][0].mbox
2
>>> for i, cmd in enumerate(mbox.cmd_log):
3
...     print(f"[{cmd.timestamp:>16}] opcode={cmd.opcode:#06x} ret={cmd.ret}")
4
[1719724823501] opcode=0x4400 ret=0      # Get Health Info OK
5
[1719724824112] opcode=0x4300 ret=0      # Get LSA OK
6
[1719724824890] opcode=0x4302 ret=-110   # Set LSA TIMEOUT ← 의심
7
[1719724825101] opcode=0x0000 ret=-19    # device removed

1
>>> # 모든 노드 정보
2
>>> for node in prog["node_data"]:
3
...     if node:
4
...         print(f"node {node.node_id}: ", end="")
5
...         print(f"present={node.node_present_pages}, ", end="")
6
...         print(f"online={node.node_spanned_pages}")
7
node 0: present=33554432, online=33554432    # DDR socket 0
8
node 1: present=33554432, online=33554432    # DDR socket 1
9
node 2: present=67108864, online=0           # CXL — offline at crash!