lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  PHC 
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:	Mon, 9 Nov 2009 16:34:20 +0900
From:	KAMEZAWA Hiroyuki <>
Cc:	"" <>,
	"" <>,
	"" <>,,
	"" <>
Subject: Re: [PATCH 0/2] memcg make use of new percpu implementations

On Mon, 9 Nov 2009 12:34:21 +0530
Balbir Singh <> wrote:

> * KAMEZAWA Hiroyuki <> [2009-11-06 17:52:42]:
> > Hi,
> > 
> > Recent updates on dynamic percpu allocation looks good and I tries to rewrite
> > memcg's poor implementation of percpu status counter.
> > (It's not NUMA-aware ...)
> > Thanks for great works.
> > 
> > For this time. I added Christoph to CC because I'm not fully sure my usage of
> > __this_cpu_xxx is correct...I'm glad if you check the usage when you have time.
> > 
> > 
> > Patch 1/2 is just clean up (prepare for patch 2/2)
> > Patch 2/2 is for percpu.
> > 
> > Tested on my 8cpu box and works well.
> > Pathcesa are against the latest mmotm.
> How do the test results look? DO you see a significant boost? 

Because my test enviroment is just an SMP (not NUMA), improvement will
not be siginificant (visible), I think.

But It's good to make use of recent updates of per_cpu implementation. We
can make use of offset calculation methods of them and one-instruction-access

This is code size (PREEMPT=y)

[Before patch]
[kamezawa@...extal mmotm-2.6.32-Nov2]$ size mm/memcontrol.o
   text    data     bss     dec     hex filename
  22403    3420    4132   29955    7503 mm/memcontrol.o

[After patch]
[kamezawa@...extal mmotm-2.6.32-Nov2]$ size mm/memcontrol.o
   text    data     bss     dec     hex filename
  22188    3420    4132   29740    742c mm/memcontrol.o

Then, text size is surely reduced.

One example is mem_cgroup_swap_statistics(). This function is inlined
after this patch.

this code: mem_cgroup_uncharge_swap(), modifies percpu counter.
        memcg = mem_cgroup_lookup(id);
        if (memcg) {
                 * We uncharge this because swap is freed.
                 * This memcg can be obsolete one. We avoid calling css_tryget
                if (!mem_cgroup_is_root(memcg))
                        res_counter_uncharge(&memcg->memsw, PAGE_SIZE);
                mem_cgroup_swap_statistics(memcg, false);

[before patch]
mem_cgroup_swap_statistics() is not inlined and uses 0x69 bytes.

0000000000001d30 <mem_cgroup_swap_statistics>:
    1d30:       push   %rbp
    1d31:       mov    %rsp,%rbp
    1d34:       push   %r12
    1d88:       callq  1d8d <mem_cgroup_swap_statistics+0x5d>
    1d8d:       nopl   (%rax)
    1d90:       jmp    1d83 <mem_cgroup_swap_statistics+0x53>
    1d92:       nopw   %cs:0x0(%rax,%rax,1)

[After patch] 
mem_cgroup_uncharge_swap()'s inlined code.

    3b67:       cmp    0x0(%rip),%rax        # 3b6e <mem_cgroup_uncharge_swap+0
    3b6e:       je     3b81 <mem_cgroup_uncharge_swap+0xd1>   <=== check mem_cgroup_is_root
    3b70:       lea    0x90(%rax),%rdi
    3b77:       mov    $0x1000,%esi
    3b7c:       callq  3b81 <mem_cgroup_uncharge_swap+0xd1>  <===  calling res_counter_uncahrge()
    3b81:       48 89 df                mov    %rbx,%rdi
    3b84:       48 8b 83 70 01 00 00    mov    0x170(%rbx),%rax   <=== get offset of mem->cpustat
    3b8b:       65 48 83 40 30 ff       addq   $0xffffffffffffffff,%gs:0x30(%rax)  mem->cpustat.count[index]--;
    3b91:       e8 6a e0 ff ff          callq  1c00 <mem_cgroup_put>

This uses 2 instruction.

Then, code size reduction is enough large, I think.

>BTW, I've been experimenting a bit with the earlier percpu counter patches,
> I might post an iteration once I have some good results.
Thank you, it's helpful.


To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

Powered by blists - more mailing lists