查看 Go 内存分配，用 strace 跟踪 mmap 调用的数据和 pmap 看到的内存不一致？

[xing393939]

1 ，test 代码见： https://go.dev/play/p/2xa_tqKsjWH

2 ，使用命令 strace -e 'signal=!all' -e '!rt_sigreturn,rt_sigaction,rt_sigprocmask,futex' ./test 得到

execve("./test", ["./test"], 0x7fff95cd9970 /* 21 vars */) = 0
arch_prctl(ARCH_SET_FS, 0x4cba70)       = 0
sched_getaffinity(0, 8192, [0, 1, 2, 3, 4, 5, 6, 7]) = 32
openat(AT_FDCWD, "/sys/kernel/mm/transparent_hugepage/hpage_pmd_size", O_RDONLY) = 3
read(3, "2097152\n", 20)                = 8
close(3)                                = 0
mmap(NULL, 262144, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f0b9870f000
mmap(NULL, 131072, PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f0b986ef000
mmap(NULL, 1048576, PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f0b985ef000
mmap(NULL, 8388608, PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f0b97def000
mmap(NULL, 67108864, PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f0b93def000
mmap(NULL, 536870912, PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f0b73def000
mmap(0xc000000000, 67108864, PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xc000000000
mmap(0xc000000000, 67108864, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0) = 0xc000000000
mmap(NULL, 33554432, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f0b71def000
mmap(NULL, 2165776, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f0b71bde000
mmap(0x7f0b986ef000, 131072, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0) = 0x7f0b986ef000
mmap(0x7f0b9866f000, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0) = 0x7f0b9866f000
mmap(0x7f0b981f5000, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0) = 0x7f0b981f5000
mmap(0x7f0b95e1f000, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0) = 0x7f0b95e1f000
mmap(0x7f0b83f6f000, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0) = 0x7f0b83f6f000
mmap(NULL, 1048576, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f0b71ade000
mmap(NULL, 65536, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f0b71ace000
mmap(NULL, 65536, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f0b71abe000
sigaltstack(NULL, {ss_sp=NULL, ss_flags=SS_DISABLE, ss_size=0}) = 0
sigaltstack({ss_sp=0xc000002000, ss_flags=0, ss_size=32768}, NULL) = 0
gettid()                                = 3689
clone(child_stack=0xc000050000, flags=CLONE_VM|CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|CLONE_SYSVSEM|CLONE_SETTLS, tls=0xc000040090) = 3690
clone(child_stack=0xc000052000, flags=CLONE_VM|CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|CLONE_SYSVSEM|CLONE_SETTLS, tls=0xc000040490) = 3691
clone(child_stack=0xc00004c000, flags=CLONE_VM|CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|CLONE_SYSVSEM|CLONE_SETTLS, tls=0xc000040890) = 3693
write(2, "0xc00003c768", 120xc00003c768)            = 12

3 ，使用pmap -p 3689得到

0000000000400000    380K r-x-- /root/logs/test
000000000045f000    420K r---- /root/logs/test
00000000004c8000     16K rw--- /root/logs/test
00000000004cc000    200K rw---   [ anon ]
000000c000000000  65536K rw---   [ anon ]
00007f0b71a7e000  36292K rw---   [ anon ]
00007f0b73def000 263680K -----   [ anon ]
00007f0b83f6f000      4K rw---   [ anon ]
00007f0b83f70000 293564K -----   [ anon ]
00007f0b95e1f000      4K rw---   [ anon ]
00007f0b95e20000  36692K -----   [ anon ]
00007f0b981f5000      4K rw---   [ anon ]
00007f0b981f6000   4580K -----   [ anon ]
00007f0b9866f000      4K rw---   [ anon ]
00007f0b98670000    508K -----   [ anon ]
00007f0b986ef000    384K rw---   [ anon ]
00007ffdc654c000    132K rw---   [ stack ]
00007ffdc65a0000     12K r----   [ anon ]
00007ffdc65a3000      8K r-x--   [ anon ]

我发现 2 个令人疑惑的地方，希望有大佬能指点：

1. 地址 7f0b986ef000 处，mmap 调用明明只申请了 128K 内存，怎么 pmap 命令却显示此处有 384K 内存。
2. pmap 显示地址 7f0b71a7e000 处的内存有 36292K ，这个内存地址是如何产生的？ starce 跟踪到的 mmap 调用并没有看到这个地址。

感谢大家的指定，2 点疑惑已经解除：
1. 地址 7f0b986ef000 处申请了 128k ，紧接着的地址 7f0b9870f000 又申请了 256k ，而在 pmap 的视图上把这 2 块内存合并了
2. 7f0b71a7e000 的地址，在我加上-f 跟踪子线程后，发现子线程又调用了 mmap ，正好是这个地址

[beordle]

你没有 -f 跟踪子线程 可以先增加跟踪，看看是不是符合预期。

[katsusan]

kernel 会对进程的 memory mapping 作 merge,比如你先申请 128K 再申请 256K,如果 128K 后的 gap 里有 256K 空间可用,
这两块 vma 会被合并在到一起, 在 process mapping 上就体现为一块 384K 的虚拟地址区域.