|
| 1 | +# Copyright (c) 2024,2025, Oracle and/or its affiliates. |
| 2 | +# Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/ |
| 3 | +""" |
| 4 | +LRU Lists |
| 5 | +------------ |
| 6 | +
|
| 7 | +Helper to work with LRU lists. LRU can be created to be memcg aware and |
| 8 | +ordered by NUMA node. |
| 9 | +
|
| 10 | +The routines iterate through the specified LRU and on NUMA machines, the |
| 11 | +output keeps the entries ordered by NUMA node. |
| 12 | +
|
| 13 | +The list_lru_from_memcg_node_for_each_entry() function allows the user to |
| 14 | +restrict the iteration by the memcg index when the list_lru is memcg_aware |
| 15 | +and the NUMA node identifier. |
| 16 | +
|
| 17 | +list_lru_kmem_to_memcgidx() is a helper to find the mem_cgroup index |
| 18 | +from a list_lru kvm address. This helper will find the memcg of a list_lru |
| 19 | +address. This routine is only interested in slab allocated entries and does |
| 20 | +not check nor handle the MEMCG_DATA_KMEM case. |
| 21 | +""" |
| 22 | +from typing import Iterator |
| 23 | +from typing import Union |
| 24 | + |
| 25 | +from drgn import cast |
| 26 | +from drgn import IntegerLike |
| 27 | +from drgn import NULL |
| 28 | +from drgn import Object |
| 29 | +from drgn import Program |
| 30 | +from drgn import Type |
| 31 | +from drgn.helpers.linux.list import list_empty |
| 32 | +from drgn.helpers.linux.list import list_for_each_entry |
| 33 | +from drgn.helpers.linux.mm import compound_head |
| 34 | +from drgn.helpers.linux.mm import page_to_pfn |
| 35 | +from drgn.helpers.linux.mm import page_to_virt |
| 36 | +from drgn.helpers.linux.mm import PageSlab |
| 37 | +from drgn.helpers.linux.mm import virt_to_page |
| 38 | +from drgn.helpers.linux.nodemask import for_each_online_node |
| 39 | +from drgn.helpers.linux.nodemask import node_state |
| 40 | +from drgn.helpers.linux.xarray import xa_for_each |
| 41 | +from drgn.helpers.linux.xarray import xa_load |
| 42 | + |
| 43 | +from drgn_tools.meminfo import get_active_numa_nodes |
| 44 | +from drgn_tools.util import has_member |
| 45 | + |
| 46 | +MEMCG_DATA_OBJCGS = 1 |
| 47 | +MEMCG_DATA_KMEM = 2 |
| 48 | + |
| 49 | +__all__ = ( |
| 50 | + "list_lru_for_each_entry", |
| 51 | + "list_lru_from_memcg_node_for_each_entry", |
| 52 | + "list_lru_kmem_to_memcgidx", |
| 53 | + "list_lru_kmem_to_nodeid", |
| 54 | +) |
| 55 | + |
| 56 | + |
| 57 | +def list_lru_for_each_entry( |
| 58 | + prog: Program, type: Union[str, Type], lru: Object, member: str |
| 59 | +) -> Iterator[Object]: |
| 60 | + """ |
| 61 | + Iterate over all of the entries in a list_lru. |
| 62 | +
|
| 63 | + :param prog: Kernel being debugged |
| 64 | + :param type: Entry type. |
| 65 | + :param lru: ``struct list_lru *`` |
| 66 | + :param member: Name of list node member in entry type. |
| 67 | + :return: Iterator of ``type *`` objects. |
| 68 | + """ |
| 69 | + memcg_aware = 0 |
| 70 | + if has_member(lru, "memcg_aware") and lru.memcg_aware: |
| 71 | + memcg_aware = 1 |
| 72 | + |
| 73 | + if has_member(lru, "node"): |
| 74 | + # no lru.node in uek7 but covered in above test |
| 75 | + if has_member(lru.node, "memcg_lrus") and lru.node[0].memcg_lrus: |
| 76 | + memcg_aware = 1 |
| 77 | + |
| 78 | + if memcg_aware: |
| 79 | + if has_member(lru, "ext") or has_member(lru, "xa"): |
| 80 | + # v5.13 (uek7) or newer |
| 81 | + if has_member(lru, "ext"): |
| 82 | + # uek7 has a UEK_KABI_REPLACE of node to ext |
| 83 | + xa = lru.ext.xa |
| 84 | + else: |
| 85 | + # uek8 |
| 86 | + xa = lru.xa |
| 87 | + # Keep the entries grouped by the NUMA node. |
| 88 | + for nid in for_each_online_node(prog): |
| 89 | + for _, memcg in xa_for_each(xa.address_of_()): |
| 90 | + # convert from the void ptr |
| 91 | + memcg = Object(prog, "struct list_lru_memcg *", memcg) |
| 92 | + lru_one = memcg.node[nid] |
| 93 | + if lru_one.nr_items > 0: |
| 94 | + for entry in list_for_each_entry( |
| 95 | + type, lru_one.list.address_of_(), member |
| 96 | + ): |
| 97 | + yield entry |
| 98 | + else: |
| 99 | + # Before v5.13, memcg entries are in an array |
| 100 | + # Keep the entries grouped by the NUMA node. |
| 101 | + for nid in for_each_online_node(prog): |
| 102 | + i = 0 |
| 103 | + while i < prog["memcg_nr_cache_ids"]: |
| 104 | + li = lru.node[nid].memcg_lrus.lru[i].list |
| 105 | + i = i + 1 |
| 106 | + if not list_empty(li.address_of_()): |
| 107 | + for entry in list_for_each_entry( |
| 108 | + type, li.address_of_(), member |
| 109 | + ): |
| 110 | + yield entry |
| 111 | + else: |
| 112 | + # not lru.memcg_aware |
| 113 | + for nid in for_each_online_node(prog): |
| 114 | + # not lru.memcg_aware |
| 115 | + if has_member(lru, "ext"): |
| 116 | + li = lru.ext.node[nid].lru.list |
| 117 | + else: |
| 118 | + li = lru.node[nid].lru.list |
| 119 | + for entry in list_for_each_entry(type, li.address_of_(), member): |
| 120 | + yield entry |
| 121 | + |
| 122 | + |
| 123 | +def list_lru_from_memcg_node_for_each_entry( |
| 124 | + prog: Program, |
| 125 | + mindx: IntegerLike, |
| 126 | + nid: IntegerLike, |
| 127 | + type: Union[str, Type], |
| 128 | + lru: Object, |
| 129 | + member: str, |
| 130 | +) -> Iterator[Object]: |
| 131 | + """ |
| 132 | + Iterate over the entries in a list_lru by the provided memcg and NUMA node. |
| 133 | +
|
| 134 | + :param prog: Kernel being debugged |
| 135 | + :param mindx: memcg index. |
| 136 | + :param nid: NUMA node ID. |
| 137 | + :param type: Entry type. |
| 138 | + :param lru: ``struct list_lru *`` |
| 139 | + :param member: Name of list node member in entry type. |
| 140 | + :return: Iterator of ``type *`` objects. |
| 141 | + """ |
| 142 | + if node_state(nid, prog["N_ONLINE"]): |
| 143 | + memcg_aware = 0 |
| 144 | + if has_member(lru, "memcg_aware") and lru.memcg_aware: |
| 145 | + memcg_aware = 1 |
| 146 | + if has_member(lru, "node"): |
| 147 | + # no lru.node in uek7 but covered in above test |
| 148 | + if has_member(lru.node, "memcg_lrus") and lru.node[0].memcg_lrus: |
| 149 | + memcg_aware = 1 |
| 150 | + if memcg_aware: |
| 151 | + if has_member(lru, "ext") or has_member(lru, "xa"): |
| 152 | + # v5.13 (uek7) or newer |
| 153 | + if has_member(lru, "ext"): |
| 154 | + # uek7 has a UEK_KABI_REPLACE of node to ext |
| 155 | + xa = lru.ext.xa |
| 156 | + else: |
| 157 | + # uek8 |
| 158 | + xa = lru.xa |
| 159 | + # Keep the entries grouped by the NUMA node. |
| 160 | + memcg = xa_load(xa.address_of_(), mindx) |
| 161 | + # convert from the void ptr unless it is a NULL |
| 162 | + if memcg != NULL(prog, "void *"): |
| 163 | + memcg = Object(prog, "struct list_lru_memcg *", memcg) |
| 164 | + lru_one = memcg.node[nid] |
| 165 | + if lru_one.nr_items > 0: |
| 166 | + for entry in list_for_each_entry( |
| 167 | + type, lru_one.list.address_of_(), member |
| 168 | + ): |
| 169 | + yield entry |
| 170 | + else: |
| 171 | + # Before v5.13 |
| 172 | + # make sure the memcg index is within the legal limits |
| 173 | + if mindx >= 0 and mindx < prog["memcg_nr_cache_ids"]: |
| 174 | + li = lru.node[nid].memcg_lrus.lru[mindx].list |
| 175 | + if not list_empty(li.address_of_()): |
| 176 | + for entry in list_for_each_entry( |
| 177 | + type, li.address_of_(), member |
| 178 | + ): |
| 179 | + yield entry |
| 180 | + else: |
| 181 | + # not lru.memcg_aware |
| 182 | + if has_member(lru, "ext"): |
| 183 | + li = lru.ext.node[nid].lru.list |
| 184 | + else: |
| 185 | + li = lru.node[nid].lru.list |
| 186 | + for entry in list_for_each_entry(type, li.address_of_(), member): |
| 187 | + yield entry |
| 188 | + |
| 189 | + |
| 190 | +def list_lru_kmem_to_memcgidx(prog: Program, kvm: IntegerLike) -> IntegerLike: |
| 191 | + """ |
| 192 | + Return the memcg index of the list_lru entry. |
| 193 | + Return -1 if the list_lru is not memcg enabled or value could not be found. |
| 194 | + Memory cgroups for slab allocation are per object. This code expects a slab |
| 195 | + allocated kvm and the MEMCG_DATA_KMEM case is NOT covered in this routine. |
| 196 | +
|
| 197 | + :param prog: Kernel being debugged |
| 198 | + :param kvm: address of a list_lru |
| 199 | + :return: memcg index, -1 means not found |
| 200 | + """ |
| 201 | + page = virt_to_page(prog, kvm) |
| 202 | + cpage = compound_head(page) |
| 203 | + # page_objcgs_check() MEMCG_DATA_OBJCGS memcg are managed per object |
| 204 | + if has_member(cpage, "memcg_data") or has_member(cpage, "obj_cgroups"): |
| 205 | + if has_member(cpage, "memcg_data"): |
| 206 | + memcg_data = cpage.memcg_data |
| 207 | + else: |
| 208 | + # cast to an integer for the MEMCG_DATA_KMEM test. |
| 209 | + memcg_data = cast("unsigned long", cpage.obj_cgroups) |
| 210 | + if memcg_data & MEMCG_DATA_OBJCGS: |
| 211 | + objcgrp = Object( |
| 212 | + prog, "struct obj_cgroup **", memcg_data - MEMCG_DATA_OBJCGS |
| 213 | + ) |
| 214 | + # offset of object calculation |
| 215 | + pvm = page_to_virt(cpage) |
| 216 | + kvm = Object(prog, "void *", kvm) |
| 217 | + if has_member(cpage, "slab_cache"): |
| 218 | + slab_cache = cpage.slab_cache |
| 219 | + else: |
| 220 | + # v5.17 (uek8) moved the kmem_cache to a new slab structure. |
| 221 | + # and since v6.10 the slab pages are identified by a page type |
| 222 | + if PageSlab(cpage): |
| 223 | + slab = Object(prog, "struct slab *", cpage) |
| 224 | + slab_cache = slab.slab_cache |
| 225 | + else: |
| 226 | + return -1 |
| 227 | + objoffset = (kvm - pvm) / slab_cache.size |
| 228 | + memcgrp = objcgrp[objoffset].memcg |
| 229 | + if memcgrp == NULL(prog, "struct mem_cgroup *"): |
| 230 | + return -1 |
| 231 | + else: |
| 232 | + return memcgrp.kmemcg_id |
| 233 | + else: |
| 234 | + return -1 |
| 235 | + else: |
| 236 | + # Before v5.13 |
| 237 | + scache = cpage.slab_cache |
| 238 | + if scache == NULL(prog, "struct kmem_cache *"): |
| 239 | + return -1 |
| 240 | + else: |
| 241 | + return cpage.slab_cache.memcg_params.memcg.kmemcg_id |
| 242 | + |
| 243 | + |
| 244 | +def list_lru_kmem_to_nodeid(prog: Program, kvm: IntegerLike) -> IntegerLike: |
| 245 | + """ |
| 246 | + Return the NUMA node id of the list_lru entry. |
| 247 | +
|
| 248 | + :param prog: Kernel being debugged |
| 249 | + :param kvm: address of a list_lru entry |
| 250 | + :return: NUMA node id |
| 251 | + """ |
| 252 | + page = virt_to_page(prog, kvm) |
| 253 | + cpage = compound_head(page) |
| 254 | + # |
| 255 | + # Stephen Brennan routine to convert a page to node |
| 256 | + pfn = page_to_pfn(cpage) |
| 257 | + nodes = get_active_numa_nodes(cpage.prog_) |
| 258 | + for i in range(1, len(nodes)): |
| 259 | + if nodes[i - 1].node_start_pfn <= pfn < nodes[i].node_start_pfn: |
| 260 | + return nodes[i - 1].node_id |
| 261 | + return nodes[-1].node_id |
0 commit comments