gdb: optimize memory commands

Tested on a complex project, results are promising.

Command                 Time cost(s)           Time saving(s)   Peformance Boost
                        Before      After
memleak                 39.494172   22.366415   17.127757       1.8
memdump                 41.872441   26.458386   15.414055       1.6
memdump -a 0x1234       28.116294   1.114119    27.002175       25.2
memdump --no-backtrace    N/A       1.114119
memmap                  7.973809    6.836468    1.137341        1.2

Signed-off-by: Xu Xingliang <xuxingliang@xiaomi.com>
This commit is contained in:
Xu Xingliang 2024-11-01 10:29:04 +08:00 committed by Xiang Xiao
parent 9f9cc7eceb
commit 412644fcf2
5 changed files with 1203 additions and 894 deletions

File diff suppressed because it is too large Load diff

637
tools/gdb/nuttxgdb/mm.py Normal file
View file

@ -0,0 +1,637 @@
############################################################################
# tools/gdb/nuttxgdb/mm.py
#
# Licensed to the Apache Software Foundation (ASF) under one or more
# contributor license agreements. See the NOTICE file distributed with
# this work for additional information regarding copyright ownership. The
# ASF licenses this file to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance with the
# License. You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
# License for the specific language governing permissions and limitations
# under the License.
#
############################################################################
from __future__ import annotations
import argparse
from typing import Generator, List, Tuple
import gdb
from . import lists, utils
from .protocols import mm as p
from .utils import Value
CONFIG_MM_BACKTRACE = utils.get_symbol_value("CONFIG_MM_BACKTRACE")
CONFIG_MM_BACKTRACE = -1 if CONFIG_MM_BACKTRACE is None else int(CONFIG_MM_BACKTRACE)
PID_MM_INVALID = -100
PID_MM_MEMPOOL = -1
class MemPoolBlock:
"""
Memory pool block instance.
"""
MAGIC_ALLOC = 0x5555_5555
mempool_backtrace_s = utils.lookup_type("struct mempool_backtrace_s")
def __init__(self, addr: int, blocksize: int, overhead: int) -> None:
"""
Initialize the memory pool block instance.
block: must be start address of the block,
blocksize: block size without backtrace overhead,
overhead: backtrace overhead size.
"""
self.overhead = overhead
self.from_pool = True
self.is_orphan = False
self.address = addr
self.blocksize = int(blocksize)
self.nodesize = int(blocksize) + self.overhead
# Lazy evaluation
self._backtrace = self._pid = self._seqno = self._magic = self._blk = None
def __repr__(self) -> str:
return f"block@{hex(self.address)},size:{self.blocksize},seqno:{self.seqno},pid:{self.pid}"
def __str__(self) -> str:
return self.__repr__()
def __hash__(self) -> int:
return hash((self.pid, self.nodesize, self.backtrace))
def __eq__(self, value: MemPoolBlock) -> bool:
return (
self.pid == value.pid
and self.nodesize == value.nodesize
and self.backtrace == value.backtrace
)
def contains(self, address: int) -> bool:
"""Check if the address is in block's range, excluding overhead"""
return self.address <= address < self.address + self.blocksize
@property
def blk(self) -> p.MemPoolBlock:
if not self._blk:
addr = int(self.address) + self.blocksize
self._blk = (
gdb.Value(addr).cast(self.mempool_backtrace_s.pointer()).dereference()
)
return self._blk
@property
def is_free(self) -> bool:
if not self._magic:
self._magic = int(self.blk["magic"])
return CONFIG_MM_BACKTRACE and self._magic != self.MAGIC_ALLOC
@property
def seqno(self) -> int:
if not self._seqno:
self._seqno = int(self.blk["seqno"]) if CONFIG_MM_BACKTRACE >= 0 else -100
return self._seqno
@property
def pid(self) -> int:
if not self._pid:
self._pid = (
int(self.blk["pid"]) if CONFIG_MM_BACKTRACE >= 0 else PID_MM_INVALID
)
return self._pid
@property
def backtrace(self) -> Tuple[int]:
if CONFIG_MM_BACKTRACE <= 0:
return ()
if not self._backtrace:
self._backtrace = tuple(
int(self.blk["backtrace"][i]) for i in range(CONFIG_MM_BACKTRACE)
)
return self._backtrace
def read_memory(self) -> memoryview:
return gdb.selected_inferior().read_memory(self.address, self.blocksize)
class MemPool(Value, p.MemPool):
"""
Memory pool instance.
"""
def __init__(self, mpool: Value, name=None) -> None:
if mpool.type.code == gdb.TYPE_CODE_PTR:
mpool = mpool.dereference()
super().__init__(mpool)
self._blksize = None
self._nfree = None
self._nifree = None
self._overhead = None
def __repr__(self) -> str:
return f"{self.name}@{hex(self.address)},size:{self.size}/{self['blocksize']},nused:{self.nused},nfree:{self.nfree}"
def __str__(self) -> str:
return self.__repr__()
@property
def name(self) -> str:
try:
return self.procfs.name.string()
except Exception:
return "<noname>"
@property
def memranges(self) -> Generator[Tuple[int, int], None, None]:
"""Memory ranges of the pool"""
sq_entry_t = utils.lookup_type("sq_entry_t")
blksize = self.size
if self.ibase:
blks = int(self.interruptsize) // blksize
base = int(self.ibase)
yield (base, base + blks * blksize)
if not self.equeue.head:
return None
# First queue has size of initialsize
ninit = int(self.initialsize)
ninit = ninit and (ninit - sq_entry_t.sizeof) // blksize
nexpand = (int(self.expandsize) - sq_entry_t.sizeof) // blksize
for entry in lists.NxSQueue(self.equeue):
blks = ninit or nexpand
ninit = 0
yield (int(entry) - blks * blksize, int(entry))
@property
def size(self) -> int:
"""Real block size including backtrace overhead"""
if not self._blksize:
blksize = self["blocksize"]
backtrace = utils.get_symbol_value("CONFIG_MM_BACKTRACE")
if CONFIG_MM_BACKTRACE is not None and backtrace >= 0:
mempool_backtrace_s = utils.lookup_type("struct mempool_backtrace_s")
size_t = utils.lookup_type("size_t")
align = (
utils.get_symbol_value("CONFIG_MM_DEFAULT_ALIGNMENT")
or 2 * size_t.sizeof
)
blksize = blksize + mempool_backtrace_s.sizeof
blksize = (blksize + align - 1) & ~(align - 1)
self._blksize = int(blksize)
return self._blksize
@property
def overhead(self) -> int:
if not self._overhead:
self._overhead = self.size - int(self["blocksize"])
return self._overhead
@property
def nwaiter(self) -> int:
return -int(self.waitsem.semcount) if self.wait and self.expandsize == 0 else 0
@property
def nused(self) -> int:
return int(self.nalloc)
@property
def free(self) -> int:
return (self.nfree + self.nifree) * self.size
@property
def nfree(self) -> int:
if not self._nfree:
self._nfree = lists.sq_count(self.queue)
return self._nfree + self.nifree
@property
def nifree(self) -> int:
"""Interrupt pool free blocks count"""
if not self._nifree:
self._nifree = lists.sq_count(self.iqueue)
return self._nifree
@property
def total(self) -> int:
nqueue = lists.sq_count(self.equeue)
sq_entry_t = utils.lookup_type("sq_entry_t")
blocks = self.nused + self.nfree
return int(nqueue * sq_entry_t.sizeof + blocks * self.size)
@property
def blks(self) -> Generator[MemPoolBlock, None, None]:
"""Iterate over all blocks in the pool"""
sq_entry_t = utils.lookup_type("sq_entry_t")
blksize = self.size # Real block size including backtrace overhead
blocksize = self["blocksize"]
def iterate(entry, nblocks):
base = int(entry) - nblocks * blksize
while nblocks > 0:
yield MemPoolBlock(base, blocksize, self.overhead)
base += blksize
nblocks -= 1
if self.ibase:
blks = int(self.interruptsize) // blksize
yield from iterate(self.ibase + blks * blksize, blks)
if not self.equeue.head:
return None
# First queue has size of initialsize
ninit = int(self.initialsize)
ninit = ninit and (ninit - sq_entry_t.sizeof) // blksize
nexpand = (int(self.expandsize) - sq_entry_t.sizeof) // blksize
for entry in lists.NxSQueue(self.equeue):
yield from iterate(entry, ninit or nexpand)
ninit = 0
def contains(self, address: int) -> Tuple[bool, Value]:
ranges = self.memranges
if not ranges:
return False, None
for start, end in ranges:
if start <= address < end:
return True, None
def find(self, address: int) -> Value:
"""Find the block that contains the given address"""
sq_entry_t = utils.lookup_type("sq_entry_t")
blksize = self.size
blocksize = self["blocksize"]
def get_blk(base):
blkstart = base + (address - base) // blksize * blksize
return MemPoolBlock(blkstart, blocksize, self.overhead)
if self.ibase:
# Check if it belongs to interrupt pool
blks = int(self.interruptsize) // blksize
base = int(self.ibase)
if base <= address < base + blks * blksize:
return get_blk(base)
if not self.equeue.head:
return None
# First queue has size of initialsize
ninit = int(self.initialsize)
ninit = ninit and (ninit - sq_entry_t.sizeof) // blksize
nexpand = (int(self.expandsize) - sq_entry_t.sizeof) // blksize
for entry in lists.NxSQueue(self.equeue):
blks = ninit or nexpand
ninit = 0
base = int(entry) - blks * blksize
if base <= address < int(entry):
return get_blk(base)
def blks_free(self) -> Generator[MemPoolBlock, None, None]:
"""Iterate over all free blocks in the pool"""
blocksize = self["blocksize"]
for entry in lists.NxSQueue(self.queue):
yield MemPoolBlock(int(entry), blocksize, self.overhead)
def blks_used(self) -> Generator[MemPoolBlock, None, None]:
"""Iterate over all used blocks in the pool"""
return filter(lambda blk: not blk.is_free, self.blks)
class MemPoolMultiple(Value, p.MemPoolMultiple):
"""
Multiple level memory pool instance.
"""
def __init__(self, mpool: Value, name=None) -> None:
if mpool.type.code == gdb.TYPE_CODE_PTR:
mpool = mpool.dereference()
super().__init__(mpool)
def __repr__(self) -> str:
return f"Multiple Level Memory Pool: {self.address}"
def __str__(self) -> str:
return self.__repr__()
@property
def pools(self) -> Generator[MemPool, None, None]:
for pool in utils.ArrayIterator(self["pools"], self.npools):
yield MemPool(pool)
@property
def free(self) -> int:
return sum(pool.free for pool in self.pools)
class MMNode(gdb.Value, p.MMFreeNode):
"""
One memory node in the memory manager heap, either free or allocated.
The instance is always dereferenced to the actual node.
"""
MM_ALLOC_BIT = 0x1
MM_PREVFREE_BIT = 0x2
MM_MASK_BIT = MM_ALLOC_BIT | MM_PREVFREE_BIT
MM_SIZEOF_ALLOCNODE = utils.sizeof("struct mm_allocnode_s")
MM_ALLOCNODE_OVERHEAD = MM_SIZEOF_ALLOCNODE - utils.sizeof("mmsize_t")
def __init__(self, node: gdb.Value):
if node.type.code == gdb.TYPE_CODE_PTR:
node = node.dereference()
self._backtrace = None
self._address = None
self._nodesize = None
super().__init__(node)
def __repr__(self):
return (
f"{hex(self.address)}({'F' if self.is_free else 'A'}{'F' if self.is_prev_free else 'A'})"
f" size:{self.nodesize}/{self.prevsize if self.is_prev_free else '-'}"
f" seq:{self.seqno} pid:{self.pid} "
)
def __str__(self) -> str:
return self.__repr__()
def __hash__(self) -> int:
return hash((self.pid, self.nodesize, self.backtrace))
def __eq__(self, value: MMNode) -> bool:
return (
self.pid == value.pid
and self.nodesize == value.nodesize
and self.backtrace == value.backtrace
)
def contains(self, address):
"""Check if the address is in node's range, excluding oeprhead"""
return (
self.address + self.overhead
<= address
< self.address + self.nodesize - MMNode.MM_ALLOCNODE_OVERHEAD
)
def read_memory(self):
addr = int(self.address) + MMNode.MM_ALLOCNODE_OVERHEAD
size = self.nodesize - MMNode.MM_ALLOCNODE_OVERHEAD
return gdb.selected_inferior().read_memory(addr, size)
@property
def address(self) -> int:
"""Change 'void *' to int"""
if not self._address:
self._address = int(super().address)
return self._address
@property
def prevsize(self) -> int:
"""Size of preceding chunk size"""
return int(self["preceding"]) & ~MMNode.MM_MASK_BIT
@property
def nodesize(self) -> int:
"""Size of this chunk, including overhead"""
if not self._nodesize:
self._nodesize = int(self["size"]) & ~MMNode.MM_MASK_BIT
return self._nodesize
@property
def usersize(self) -> int:
"""Size of this chunk, excluding overhead"""
return self.nodesize - MMNode.MM_ALLOCNODE_OVERHEAD
@property
def flink(self):
# Only free node has flink and blink
return self["flink"] if self.is_free else None
@property
def blink(self):
# Only free node has flink and blink
return self["blink"] if self.is_free else None
@property
def pid(self) -> int:
# Only available when CONFIG_MM_BACKTRACE >= 0
if CONFIG_MM_BACKTRACE >= 0:
return int(self["pid"])
return PID_MM_INVALID
@property
def seqno(self) -> int:
return int(self["seqno"]) if CONFIG_MM_BACKTRACE >= 0 else -1
@property
def backtrace(self) -> List[Tuple[int, str, str]]:
if CONFIG_MM_BACKTRACE <= 0:
return ()
if not self._backtrace:
self._backtrace = tuple(
int(self["backtrace"][i]) for i in range(CONFIG_MM_BACKTRACE)
)
return self._backtrace
@property
def prevnode(self) -> MMNode:
addr = int(self.address) - self["presize"]
type = utils.lookup_type("struct mm_allocnode_s").pointer()
return MMNode(gdb.Value(addr).cast(type))
@property
def nextnode(self) -> MMNode:
addr = int(self.address) + self.nodesize
type = utils.lookup_type("struct mm_allocnode_s").pointer()
# Use gdb.Value for better performance
return MMNode(gdb.Value(addr).cast(type))
@property
def is_free(self) -> bool:
return not self["size"] & MMNode.MM_ALLOC_BIT
@property
def is_prev_free(self) -> bool:
return self["preceding"] & MMNode.MM_PREVFREE_BIT
@property
def is_orphan(self) -> bool:
# Report orphaned node and node likely to be orphaned(free-used-used-free)
return self.is_prev_free or self.nextnode.is_free
@property
def from_pool(self) -> bool:
return False
@property
def overhead(self) -> int:
return MMNode.MM_ALLOCNODE_OVERHEAD
class MMHeap(Value, p.MMHeap):
"""
One memory manager heap. It may contains multiple regions.
"""
def __init__(self, heap: Value, name=None) -> None:
if heap.type.code == gdb.TYPE_CODE_PTR:
heap = heap.dereference()
super().__init__(heap)
self.name = name or "<noname>"
def __repr__(self) -> str:
return f"{self.name}@{self.address}, {self.nregions}regions, {int(self.heapsize) / 1024 :.1f}kB"
def __str__(self) -> str:
return self.__repr__()
@property
def curused(self) -> int:
return int(self.mm_curused)
@property
def heapsize(self) -> int:
return int(self.mm_heapsize)
@property
def free(self) -> int:
return self.heapsize - self.curused
@property
def nregions(self) -> int:
return int(utils.get_field(self, "mm_nregions", default=1))
@property
def regions(self):
regions = self.nregions
for start, end in zip(
utils.ArrayIterator(self.mm_heapstart, regions),
utils.ArrayIterator(self.mm_heapend, regions),
):
yield MMNode(start), MMNode(end)
@property
def nodes(self) -> Generator[MMNode, None, None]:
for start, end in self.regions:
node = start
while node.address <= end.address:
yield node
node = node.nextnode
def nodes_free(self) -> Generator[MMNode, None, None]:
return filter(lambda node: node.is_free, self.nodes)
def nodes_used(self) -> Generator[MMNode, None, None]:
return filter(lambda node: not node.is_free, self.nodes)
def contains(self, address: int) -> bool:
return any(
start.address <= address < end.address for start, end in self.regions
)
def find(self, address: int) -> MMNode:
for node in self.nodes:
if node.address <= address < node.address + node.nodesize:
return node
def get_heaps() -> List[MMHeap]:
# parse g_procfs_meminfo to get all heaps
heaps = []
meminfo: p.ProcfsMeminfoEntry = utils.gdb_eval_or_none("g_procfs_meminfo")
if not meminfo and (heap := gdb.parse_and_eval("g_mmheap")):
heaps.append(MMHeap(heap))
while meminfo:
heaps.append(MMHeap(meminfo.heap, name=meminfo.name.string()))
meminfo = meminfo.next
return heaps
def get_pools(heaps: List[Value] = []) -> Generator[MemPool, None, None]:
for heap in heaps or get_heaps():
if not (mm_pool := heap.mm_mpool):
continue
mpool = MemPoolMultiple(mm_pool)
for pool in mpool.pools:
yield pool
class MMHeapInfo(gdb.Command):
"""Show basic heap information"""
def __init__(self):
super().__init__("mm heap", gdb.COMMAND_USER)
def invoke(self, arg: str, from_tty: bool) -> None:
for heap in get_heaps():
regions = [(start.address, end.address) for start, end in heap.regions]
gdb.write(f"{heap} - has {len(list(heap.nodes))} nodes, regions:")
gdb.write(" ".join(f"{hex(start)}~{hex(end)}" for start, end in regions))
gdb.write("\n")
class MMPoolInfo(gdb.Command):
"""Show basic heap information"""
def __init__(self):
super().__init__("mm pool", gdb.COMMAND_USER)
utils.alias("mempool", "mm pool")
def invoke(self, arg: str, from_tty: bool) -> None:
parser = argparse.ArgumentParser(description="Dump memory pool information.")
parser.add_argument(
"--heap", type=str, help="Which heap's pool to show", default=None
)
try:
args = parser.parse_args(gdb.string_to_argv(arg))
except SystemExit:
return
heaps = [gdb.parse_and_eval(args.heap)] if args.heap else get_heaps()
if not (pools := list(get_pools(heaps))):
gdb.write("No pools found.\n")
return
count = len(pools)
gdb.write(f"Total {count} pools\n")
name_max = max(len(pool.name) for pool in pools) + 11 # 11: "@0x12345678"
formatter = "{:>%d} {:>11} {:>9} {:>9} {:>9} {:>9} {:>9}\n" % name_max
head = ("", "total", "bsize", "nused", "nfree", "nifree", "nwaiter")
gdb.write(formatter.format(*head))
for pool in pools:
gdb.write(
formatter.format(
f"{pool.name}@{pool.address:#x}",
pool.total,
pool.size,
pool.nused,
pool.nfree,
pool.nifree,
pool.nwaiter,
)
)

View file

@ -28,3 +28,10 @@ class ForeachPrefix(gdb.Command):
def __init__(self):
super(ForeachPrefix, self).__init__("foreach", gdb.COMMAND_USER, prefix=True)
class MMPrefixCommand(gdb.Command):
"""Memory manager related commands prefix."""
def __init__(self):
super().__init__("mm", gdb.COMMAND_USER, prefix=True)

View file

@ -0,0 +1,120 @@
############################################################################
# tools/gdb/nuttxgdb/protocols/mm.py
#
# Licensed to the Apache Software Foundation (ASF) under one or more
# contributor license agreements. See the NOTICE file distributed with
# this work for additional information regarding copyright ownership. The
# ASF licenses this file to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance with the
# License. You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
# License for the specific language governing permissions and limitations
# under the License.
#
############################################################################
from __future__ import annotations
from typing import List
from .value import Value
class ProcfsMeminfoEntry(Value):
"""struct procfs_meminfo_entry_s"""
name: Value
heap: Value
next: ProcfsMeminfoEntry
class MMAllocNode(Value):
"""struct mm_allocnode_s"""
preceding: Value
size: Value
pid: Value
seqno: Value
backtrace: Value
class MMFreeNode(Value):
"""struct mm_freenode_s"""
preceding: Value
size: Value
pid: Value
seqno: Value
backtrace: Value
flink: MMFreeNode
blink: MMFreeNode
class MMHeap(Value):
"""struct mm_heap_s"""
mm_lock: Value
mm_heapsize: Value
mm_maxused: Value
mm_curused: Value
mm_heapstart: List[MMAllocNode]
mm_heapend: List[MMAllocNode]
mm_nregions: Value
mm_nodelist: Value
class MemPool(Value):
"""struct mempool_s"""
initialsize: Value
interruptsize: Value
expandsize: Value
wait: Value
priv: Value
alloc: Value
free: Value
check: Value
ibase: Value
queue: Value
iqueue: Value
equeue: Value
nalloc: Value
lock: Value
waitsem: Value
procfs: Value
class MemPoolMultiple(Value):
"""struct mempool_multiple_s"""
pools: List[MemPool]
npools: Value
expandsize: Value
minpoolsize: Value
arg: Value
alloc: Value
alloc_size: Value
free: Value
alloced: Value
delta: Value
lock: Value
chunk_queue: Value
chunk_size: Value
dict_used: Value
dict_col_num_log2: Value
dict_row_num: Value
dict: Value
class MemPoolBlock(Value):
"""struct mempool_backtrace_s"""
magic: Value
pid: Value
seqno: Value
backtrace: Value

View file

@ -500,6 +500,13 @@ def parse_arg(arg: str) -> Union[gdb.Value, int]:
return None
def alias(name, command):
try:
gdb.execute(f"alias {name} = {command}")
except gdb.error:
pass
def nitems(array):
array_type = array.type
element_type = array_type.target()