Update with test

2026-04-30 07:13:14 -04:00
parent b86e4f9a98
commit 2d43d7bc02
3 changed files with 313 additions and 0 deletions
--- a/build_merged_v2.py
+++ b/build_merged_v2.py
@@ -0,0 +1,92 @@
+#!/usr/bin/env python3
+"""
+Build merged GDT using corrected 5-chunk translation formula.
+
+With the old formula (group*4*CHUNK), chunk_idx=4 blocks were skipped and
+every address past group 0 was wrong. The correct formula is group*5*CHUNK
+for all chunk indices 0-4.
+
+Since all chunks are now readable, we use the primary GDT for everything and
+fall back to the backup (group 1) only when a primary entry reads as all-zeros
+(indicating it genuinely landed in a damaged sector).
+"""
+import struct
+
+CHUNK      = 128 * 512     # 64 KB
+LV_START   = 5120000 * 512
+BSIZE      = 4096
+BPG        = 32768
+GDT_ENTRY  = 64
+NUM_GROUPS = 35728
+
+
+def raw_read(virt_offset, length):
+    """Correct PERC H710 translation — 5-chunk stride, all chunks readable."""
+    result = bytearray(length)
+    pos = virt_offset
+    remaining = length
+    with open('/dev/md0', 'rb') as f:
+        while remaining > 0:
+            group     = pos // (5 * CHUNK)
+            in_group  = pos % (5 * CHUNK)
+            chunk_idx = in_group // CHUNK
+            intra     = in_group % CHUNK
+            seg_len   = min(CHUNK - intra, remaining)
+            dst_off   = pos - virt_offset
+            phys = LV_START + group * 5 * CHUNK + chunk_idx * CHUNK + intra
+            f.seek(phys)
+            data = f.read(seg_len)
+            result[dst_off:dst_off + len(data)] = data
+            pos       += seg_len
+            remaining -= seg_len
+    return bytes(result)
+
+
+print('Building merged GDT (v2 — corrected 5-chunk formula)...')
+primary_start = BSIZE                  # block 1
+backup_start  = (1 * BPG + 1) * BSIZE # group 1 backup SB location
+
+primary_gdt = raw_read(primary_start, NUM_GROUPS * GDT_ENTRY)
+backup_gdt  = raw_read(backup_start,  NUM_GROUPS * GDT_ENTRY)
+
+merged       = bytearray(NUM_GROUPS * GDT_ENTRY)
+primary_used = 0
+backup_used  = 0
+zero_both    = 0
+
+for g in range(NUM_GROUPS):
+    src_off = g * GDT_ENTRY
+    prim_e  = primary_gdt[src_off:src_off + GDT_ENTRY]
+    back_e  = backup_gdt[src_off:src_off + GDT_ENTRY]
+
+    if any(prim_e):
+        merged[src_off:src_off + GDT_ENTRY] = prim_e
+        primary_used += 1
+    elif any(back_e):
+        merged[src_off:src_off + GDT_ENTRY] = back_e
+        backup_used += 1
+    else:
+        zero_both += 1
+
+print(f'  From primary GDT : {primary_used}')
+print(f'  From backup GDT  : {backup_used}')
+print(f'  Both zero (lost) : {zero_both}')
+
+if zero_both > 0:
+    print(f'  WARNING: {zero_both} group descriptors could not be recovered')
+
+print()
+print('Sample entries:')
+for g in [0, 1, 13, 100, 1000, 35000, 35727]:
+    e   = merged[g * GDT_ENTRY:(g + 1) * GDT_ENTRY]
+    bb  = struct.unpack_from('<I', e,  0)[0]
+    ib  = struct.unpack_from('<I', e,  4)[0]
+    it  = struct.unpack_from('<I', e,  8)[0]
+    cs  = struct.unpack_from('<H', e, 30)[0]
+    src = 'primary' if any(primary_gdt[g*GDT_ENTRY:(g+1)*GDT_ENTRY]) else 'backup'
+    print(f'  Group {g:6d}: bb={bb:10d} ib={ib:10d} it={it:10d} '
+          f'csum=0x{cs:04x}  [{src}]')
+
+with open('/tmp/merged_gdt.bin', 'wb') as f:
+    f.write(merged)
+print(f'\nSaved /tmp/merged_gdt.bin ({len(merged) // 1024} KB)')