contrib: add script to dump page_owner info

Sample output:

Page allocated via order 0, gfp_mask: 0x140cca, pid: 74, tgid: 74 (b'kworker/u32:2'), ts 1189257596 ns, free_ts 0 ns
PFN: 262203, Flags: 0x3fffe000004003c
#0  set_page_owner (./include/linux/page_owner.h:32:3)
osandov#1  post_alloc_hook (mm/page_alloc.c:1502:2)
osandov#2  prep_new_page (mm/page_alloc.c:1510:2)
osandov#3  get_page_from_freelist (mm/page_alloc.c:3489:4)
osandov#4  __alloc_pages_noprof (mm/page_alloc.c:4747:9)
osandov#5  alloc_pages_mpol_noprof (mm/mempolicy.c:2263:9)
osandov#6  folio_alloc_mpol_noprof (mm/mempolicy.c:2281:9)
osandov#7  shmem_alloc_folio (mm/shmem.c:1726:10)
osandov#8  shmem_alloc_and_add_folio (mm/shmem.c:1786:11)
osandov#9  shmem_get_folio_gfp (mm/shmem.c:2192:10)
osandov#10 shmem_get_folio (mm/shmem.c:2297:9)
osandov#11 shmem_write_begin (mm/shmem.c:2902:8)
osandov#12 generic_perform_write (mm/filemap.c:4019:12)
osandov#13 shmem_file_write_iter (mm/shmem.c:3078:8)
osandov#14 __kernel_write_iter (fs/read_write.c:523:8)
osandov#15 __kernel_write (fs/read_write.c:543:9)
osandov#16 kernel_write (fs/read_write.c:564:9)
osandov#17 kernel_write (fs/read_write.c:554:9)
osandov#18 xwrite (init/initramfs.c:33:16)
osandov#19 do_copy (init/initramfs.c:405:7)
osandov#20 write_buffer (init/initramfs.c:452:10)
osandov#21 unpack_to_rootfs (init/initramfs.c:505:14)
...

Signed-off-by: Kuan-Ying Lee <kuan-ying.lee@canonical.com>

contrib/page_owner.py

@@ -0,0 +1,176 @@
+#!/usr/bin/env drgn
+# Copyright (c) Canonical Ltd.
+# SPDX-License-Identifier: LGPL-2.1-or-later
+
+""" Script to dump vmallocinfo status using drgn"""
+
+import re
+from typing import Optional
+
+from drgn import IntegerLike, Program, sizeof
+from drgn.helpers.linux.mm import *
+from drgn.helpers.linux.kconfig import get_kconfig
+from drgn.helpers.linux.stackdepot import stack_depot_fetch
+
+def DIV_ROUND_UP(n,d):
+    return ((n) + (d) - 1) // (d)
+
+try:
+    MAX_ORDER_NR_PAGES = int(get_kconfig(prog)["CONFIG_ARCH_FORCE_MAX_ORDER"])
+except:
+    MAX_ORDER_NR_PAGES = 10
+
+vmcoreinfo = prog["VMCOREINFO"].string_().decode()
+def get_vmcoreinfo_number(item) -> IntegerLike:
+    match = re.search(
+        r"^NUMBER\(%s\)=([0-9]+)$" % item, vmcoreinfo, flags=re.M
+    )
+    if match:
+        return int(match.group(1))
+    else:
+        raise Exception("Cannot find %s in vmcoreinfo" % item)
+
+SECTION_SIZE_BITS = get_vmcoreinfo_number("SECTION_SIZE_BITS")
+MAX_PHYSMEM_BITS = get_vmcoreinfo_number("MAX_PHYSMEM_BITS")
+SECTIONS_SHIFT = MAX_PHYSMEM_BITS - SECTION_SIZE_BITS
+
+NR_MEM_SECTIONS = 1 << SECTIONS_SHIFT
+PFN_SECTION_SHIFT = SECTION_SIZE_BITS - prog["PAGE_SHIFT"]
+
+if get_kconfig(prog)["CONFIG_SPARSEMEM_EXTREME"]:
+    SECTIONS_PER_ROOT = prog["PAGE_SIZE"].value_() // sizeof(prog.type('struct mem_section'))
+else:
+    SECTIONS_PER_ROOT = 1
+
+NR_SECTION_ROOTS = DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
+SECTION_ROOT_MASK = SECTIONS_PER_ROOT - 1
+
+SUBSECTION_SHIFT = 21
+PFN_SUBSECTION_SHIFT = SUBSECTION_SHIFT - prog["PAGE_SHIFT"]
+PAGES_PER_SUBSECTION = 1 << PFN_SUBSECTION_SHIFT
+
+PAGES_PER_SECTION = 1 << PFN_SECTION_SHIFT
+PAGE_SECTION_MASK = ~(PAGES_PER_SECTION - 1)
+
+SECTION_HAS_MEM_MAP = 1 << prog["SECTION_HAS_MEM_MAP_BIT"].value_()
+SECTION_IS_EARLY = 1 << prog["SECTION_IS_EARLY_BIT"].value_()
+
+PAGE_EXT_OWNER = prog["PAGE_EXT_OWNER"].value_()
+PAGE_EXT_OWNER_ALLOCATED = prog["PAGE_EXT_OWNER_ALLOCATED"].value_()
+
+PAGE_EXT_INVALID = 1
+
+def pfn_to_section_nr(pfn: Object) -> IntegerLike:
+    return pfn >> PFN_SECTION_SHIFT
+
+def section_nr_to_root(sec: IntegerLike) -> IntegerLike:
+    return sec.value_() // SECTIONS_PER_ROOT
+
+def nr_to_section(nr: IntegerLike) -> Object:
+    root = section_nr_to_root(nr)
+    if root >= NR_SECTION_ROOTS:
+        raise Exception("root >= NR_SECTION_ROOTS")
+    return prog["mem_section"][root][nr & SECTION_ROOT_MASK]
+
+def valid_section(section: Object) -> bool:
+    return bool(section.address_of_() and (section.section_mem_map & SECTION_HAS_MEM_MAP))
+
+def early_section(section: Object) -> bool:
+    return bool(section.address_of_() and (section.section_mem_map & SECTION_IS_EARLY))
+
+def subsection_map_index(pfn: Object) -> IntegerLike:
+    return (pfn & ~(PAGE_SECTION_MASK)) // PAGES_PER_SUBSECTION
+
+def pfn_section_valid(ms: Object, pfn: Object) -> IntegerLike:
+    idx = subsection_map_index(pfn)
+    usage = ms.usage
+    if usage:
+        return (1 << idx) & usage.subsection_map
+    else:
+        return 0
+
+def pfn_to_section(pfn: Object) -> Optional[Object]:
+    return nr_to_section(pfn_to_section_nr(pfn))
+
+def pfn_valid(pfn: Object) -> IntegerLike:
+    if PHYS_PFN(PFN_PHYS(pfn)) != pfn:
+        return 0
+    if pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS:
+        return 0
+    ms = pfn_to_section(pfn)
+    if not valid_section(ms):
+        return 0
+    ret = early_section(ms) or pfn_section_valid(ms, pfn)
+    return ret
+
+
+def page_ext_invalid(page_ext: Object) -> bool:
+    if not page_ext:
+        return True
+    if page_ext.value_() & PAGE_EXT_INVALID == PAGE_EXT_INVALID:
+        return True
+    return False
+
+def get_entry(base: Object, index: Object) -> Object:
+    return Object(prog, "struct page_ext *", (cast("unsigned long", base) + prog["page_ext_size"].value_() * index))
+
+def lookup_page_ext(page: Object) -> Object:
+    pfn = page_to_pfn(page)
+    section = pfn_to_section(pfn)
+    page_ext = section.page_ext
+    if page_ext_invalid(page_ext):
+        return drgn.NULL(prog, "unsigned long")
+    return get_entry(page_ext, pfn)
+
+def page_ext_get(page: Object) -> Optional[Object]:
+    page_ext = lookup_page_ext(page)
+    if page_ext:
+        return page_ext
+
+def get_page_owner(page_ext: Object) -> Object:
+    addr = cast("unsigned long", page_ext) + prog["page_owner_ops"].offset
+    return Object(prog, "struct page_owner *", addr)
+
+min_low_pfn = prog["min_low_pfn"]
+max_pfn = prog["max_pfn"]
+
+def read_page_owner():
+    if prog["page_owner_inited"].key.enabled.counter != 1:
+        raise Exception("page_owner is not enabled")
+    pfn = min_low_pfn
+    while (not pfn_valid(pfn)) and (pfn & (MAX_ORDER_NR_PAGES - 1) != 0):
+        pfn += 1
+    while pfn < max_pfn:
+        #
+        # If the new page is in a new MAX_ORDER_NR_PAGES area,
+        # validate the area as existing, skip it if not
+        #
+        if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) and (not pfn_valid(pfn)):
+            pfn += (MAX_ORDER_NR_PAGES - 1)
+            continue;
+
+        page = pfn_to_page(pfn)
+        page_ext = page_ext_get(page)
+        if not page_ext:
+            pfn += 1
+            continue
+
+        if not (page_ext.flags & (1 << PAGE_EXT_OWNER)):
+            pfn += 1
+            continue
+
+        if not (page_ext.flags & (1 << PAGE_EXT_OWNER_ALLOCATED)):
+            pfn += 1
+            continue
+
+        page_owner = get_page_owner(page_ext)
+        trace = stack_depot_fetch(page_owner.handle)
+        print("Page allocated via order %d, gfp_mask: 0x%x, pid: %d, tgid: %d (%s), ts %u ns, free_ts %u ns" %\
+                (page_owner.order, page_owner.gfp_mask,\
+                page_owner.pid, page_owner.tgid, page_owner.comm.string_(),\
+                page_owner.ts_nsec, page_owner.free_ts_nsec))
+        print("PFN: %d, Flags: 0x%x" % (pfn, page.flags))
+        print(trace)
+        pfn += (1 << page_owner.order)
+
+read_page_owner()