lists.openwall.net   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  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <201807311112492239674@zte.com.cn>
Date:   Tue, 31 Jul 2018 11:12:49 +0800 (CST)
From:   <cheng.lin130@....com.cn>
To:     <peterz@...radead.org>
Cc:     <mingo@...hat.com>, <linux-kernel@...r.kernel.org>,
        <jiang.biao2@....com.cn>, <zhong.weidong@....com.cn>,
        <tan.hu@....com.cn>
Subject: 答复: Re: [PATCH v2] sched/numa: do not balance tasks onto isolated cpus

>On Thu, Jul 26, 2018 at 04:19:08PM +0800, Cheng Lin wrote:






>> -    if (!cpumask_test_cpu(arg.dst_cpu, &arg.src_task->cpus_allowed))
>> +    if ((!cpumask_test_cpu(arg.dst_cpu, &arg.src_task->cpus_allowed))
>> +        || !housekeeping_test_cpu(arg.dst_cpu, HK_FLAG_DOMAIN))
>>          goto out;
>
>You did not read the comment I provided last time. Using isolcpus (and
>thus it's renamed houskeeping thing) is the wrong thing to do. Load
>balancing should be limited to it's root domain.









"isolcpus" is "[Deprecated - use cpusets instead]" in the lastest kernel(4.18).


Before 4.15, the option "isolcpus" is "the preferred way to isolate CPUs". 






Although it is "Deprecated",  but we also can use it in the lastest kernel. 


And now a large number of “isolcpus” are using in the old way. Whether NUMA-balancing 


should consider the “isolcpus” for forward compatibility? Or in some branches?
<div class="zcontentRow"><p>&gt;On&nbsp;Thu,&nbsp;Jul&nbsp;26,&nbsp;2018&nbsp;at&nbsp;04:19:08PM&nbsp;+0800,&nbsp;Cheng&nbsp;Lin&nbsp;wrote:<br></p><div><div class="zhistoryRow" style="display:block"><div id="zwriteHistoryContainer"><div class="control-group zhistoryPanel"><div class="zhistoryContent">&gt;&gt;&nbsp;-&nbsp;&nbsp;&nbsp;&nbsp;if&nbsp;(!cpumask_test_cpu(arg.dst_cpu,&nbsp;&amp;arg.src_task-&gt;cpus_allowed))<br>&gt;&gt;&nbsp;+&nbsp;&nbsp;&nbsp;&nbsp;if&nbsp;((!cpumask_test_cpu(arg.dst_cpu,&nbsp;&amp;arg.src_task-&gt;cpus_allowed))<br>&gt;&gt;&nbsp;+&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;||&nbsp;!housekeeping_test_cpu(arg.dst_cpu,&nbsp;HK_FLAG_DOMAIN))<br>&gt;&gt;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;goto&nbsp;out;<br>&gt;<br>&gt;You&nbsp;did&nbsp;not&nbsp;read&nbsp;the&nbsp;comment&nbsp;I&nbsp;provided&nbsp;last&nbsp;time.&nbsp;Using&nbsp;isolcpus&nbsp;(and<br>&gt;thus&nbsp;it's&nbsp;renamed&nbsp;houskeeping&nbsp;thing)&nbsp;is&nbsp;the&nbsp;wrong&nbsp;thing&nbsp;to&nbsp;do.&nbsp;Load<br>&gt;balancing&nbsp;should&nbsp;be&nbsp;limited&nbsp;to&nbsp;it's&nbsp;root&nbsp;domain.</div></div></div></div></div><p style="line-height: 21px; white-space: normal;"><br></p><p style="line-height: 21px; white-space: normal;">"isolcpus" is "[Deprecated - use cpusets instead]" in the lastest kernel(4.18).</p><p style="line-height: 21px; white-space: normal;">Before 4.15, the option "isolcpus" is "the preferred way to isolate CPUs".&nbsp;</p><p style="line-height: 21px; white-space: normal;"><br></p><p style="line-height: 21px; white-space: normal;">Although it is "Deprecated", &nbsp;but we also can use it in the lastest kernel.&nbsp;</p><p style="line-height: 21px; white-space: normal;">And now a large number of “isolcpus” are using in the old way. Whether NUMA-balancing&nbsp;</p><p style="line-height: 21px; white-space: normal;">should consider the “isolcpus” for forward compatibility? Or in some branches?</p><p><br></p></div>

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ