From: Yunsheng Lin <linyunsheng@huawei.com>
To: Peter Zijlstra <peterz@infradead.org>
Cc: Michal Hocko <mhocko@kernel.org>, <catalin.marinas@arm.com>,
<will@kernel.org>, <mingo@redhat.com>, <bp@alien8.de>,
<rth@twiddle.net>, <ink@jurassic.park.msu.ru>,
<mattst88@gmail.com>, <benh@kernel.crashing.org>,
<paulus@samba.org>, <mpe@ellerman.id.au>,
<heiko.carstens@de.ibm.com>, <gor@linux.ibm.com>,
<borntraeger@de.ibm.com>, <ysato@users.sourceforge.jp>,
<dalias@libc.org>, <davem@davemloft.net>, <ralf@linux-mips.org>,
<paul.burton@mips.com>, <jhogan@kernel.org>,
<jiaxun.yang@flygoat.com>, <chenhc@lemote.com>,
<akpm@linux-foundation.org>, <rppt@linux.ibm.com>,
<anshuman.khandual@arm.com>, <tglx@linutronix.de>, <cai@lca.pw>,
<robin.murphy@arm.com>, <linux-arm-kernel@lists.infradead.org>,
<linux-kernel@vger.kernel.org>, <hpa@zytor.com>, <x86@kernel.org>,
<dave.hansen@linux.intel.com>, <luto@kernel.org>,
<len.brown@intel.com>, <axboe@kernel.dk>, <dledford@redhat.com>,
<jeffrey.t.kirsher@intel.com>, <linux-alpha@vger.kernel.org>,
<naveen.n.rao@linux.vnet.ibm.com>, <mwb@linux.vnet.ibm.com>,
<linuxppc-dev@lists.ozlabs.org>, <linux-s390@vger.kernel.org>,
<linux-sh@vger.kernel.org>, <sparclinux@vger.kernel.org>,
<tbogendoerfer@suse.de>, <linux-mips@vger.kernel.org>,
<rafael@kernel.org>, <gregkh@linuxfoundation.org>
Subject: Re: [PATCH v6] numa: make node_to_cpumask_map() NUMA_NO_NODE aware
Date: Tue, 24 Sep 2019 19:44:28 +0800 [thread overview]
Message-ID: <8cff8350-311e-3817-0c42-b6f98de589a3@huawei.com> (raw)
In-Reply-To: <20190924112811.GK2332@hirez.programming.kicks-ass.net>
On 2019/9/24 19:28, Peter Zijlstra wrote:
> On Tue, Sep 24, 2019 at 07:07:36PM +0800, Yunsheng Lin wrote:
>> On 2019/9/24 17:25, Peter Zijlstra wrote:
>>> On Tue, Sep 24, 2019 at 09:29:50AM +0800, Yunsheng Lin wrote:
>>>> On 2019/9/24 4:34, Peter Zijlstra wrote:
>>>
>>>>> I'm saying the ACPI standard is wrong. Explain to me how it is
>>>>> physically possible to have a device without NUMA affinity in a NUMA
>>>>> system?
>>>>>
>>>>> 1) The fundamental interconnect is not uniform.
>>>>> 2) The device needs to actually be somewhere.
>>>>>
>>>>
>>>> From what I can see, NUMA_NO_NODE may make sense in the below case:
>>>>
>>>> 1) Theoretically, there would be a device that can access all the memory
>>>> uniformly and can be accessed by all cpus uniformly even in a NUMA system.
>>>> Suppose we have two nodes, and the device just sit in the middle of the
>>>> interconnect between the two nodes.
>>>>
>>>> Even we define a third node solely for the device, we may need to look at
>>>> the node distance to decide the device can be accessed uniformly.
>>>>
>>>> Or we can decide that the device can be accessed uniformly by setting
>>>> it's node to NUMA_NO_NODE.
>>>
>>> This is indeed a theoretical case; it doesn't scale. The moment you're
>>> adding multiple sockets or even board interconnects this all goes out
>>> the window.
>>>
>>> And in this case, forcing the device to either node is fine.
>>
>> Not really.
>> For packet sending and receiving, the buffer memory may be allocated
>> dynamically. Node of tx buffer memory is mainly based on the cpu
>> that is sending sending, node of rx buffer memory is mainly based on
>> the cpu the interrupt handler of the device is running on, and the
>> device' interrupt affinity is mainly based on node id of the device.
>>
>> We can bind the processes that are using the device to both nodes
>> in order to utilize memory on both nodes for packet sending.
>>
>> But for packet receiving, the node1 may not be used becuase the node
>> id of device is forced to node 0, which is the default way to bind
>> the interrupt to the cpu of the same node.
>>
>> If node_to_cpumask_map() returns all usable cpus when the device's node
>> id is NUMA_NO_NODE, then interrupt can be binded to the cpus on both nodes.
>
> s/binded/bound/
>
> Sure; the data can be allocated wherever, but the control structures are
> not dynamically allocated every time. They are persistent, and they will
> be local to some node.
>
> Anyway, are you saying this stupid corner case is actually relevant?
> Because how does it scale out? What if you have 8 sockets, with each
> socket having 2 nodes and 1 such magic device. Then returning all CPUs
> is just plain wrong.
Yes, the hardware may not scale out, but what about the virtual device?
>
>>>> 2) For many virtual deivces, such as tun or loopback netdevice, they
>>>> are also accessed uniformly by all cpus.
>>>
>>> Not true; the virtual device will sit in memory local to some node.
>>>
>>> And as with physical devices, you probably want at least one (virtual)
>>> queue per node.
>>
>> There may be similar handling as above for virtual device too.
>
> And it'd be similarly broken.
From [1], there is a lot of devices with node id of NUMA_NO_NODE with the
FW_BUG.
[1] https://lore.kernel.org/lkml/5a188e2b-6c07-a9db-fbaa-561e9362d3ba@huawei.com/
>
> .
>
next prev parent reply other threads:[~2019-09-24 11:44 UTC|newest]
Thread overview: 53+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-09-17 12:48 [PATCH v6] numa: make node_to_cpumask_map() NUMA_NO_NODE aware Yunsheng Lin
2019-09-23 15:15 ` Peter Zijlstra
2019-09-23 15:28 ` Michal Hocko
2019-09-23 15:48 ` Peter Zijlstra
2019-09-23 16:52 ` Michal Hocko
2019-09-23 20:34 ` Peter Zijlstra
2019-09-24 1:29 ` Yunsheng Lin
2019-09-24 9:25 ` Peter Zijlstra
2019-09-24 11:07 ` Yunsheng Lin
2019-09-24 11:28 ` Peter Zijlstra
2019-09-24 11:44 ` Yunsheng Lin [this message]
2019-09-24 11:58 ` Peter Zijlstra
2019-09-24 12:09 ` Yunsheng Lin
2019-09-24 7:47 ` Michal Hocko
2019-09-24 9:17 ` Peter Zijlstra
2019-09-24 10:56 ` Michal Hocko
2019-09-24 11:23 ` Peter Zijlstra
2019-09-24 11:54 ` Michal Hocko
2019-09-24 12:09 ` Peter Zijlstra
2019-09-24 12:25 ` Michal Hocko
2019-09-24 12:43 ` Peter Zijlstra
2019-09-24 12:59 ` Peter Zijlstra
2019-09-24 13:19 ` Michal Hocko
2019-09-25 9:14 ` Yunsheng Lin
2019-09-25 10:41 ` Peter Zijlstra
2019-10-08 8:38 ` Yunsheng Lin
2019-10-09 12:25 ` Robin Murphy
2019-10-10 6:07 ` Yunsheng Lin
2019-10-10 7:32 ` Michal Hocko
2019-10-11 3:27 ` Yunsheng Lin
2019-10-11 11:15 ` Peter Zijlstra
2019-10-12 6:17 ` Yunsheng Lin
2019-10-12 7:40 ` Greg KH
2019-10-12 9:47 ` Yunsheng Lin
2019-10-12 10:40 ` Greg KH
2019-10-12 10:47 ` Greg KH
2019-10-14 8:00 ` Yunsheng Lin
2019-10-14 9:25 ` Greg KH
2019-10-14 9:49 ` Peter Zijlstra
2019-10-14 10:04 ` Greg KH
2019-10-15 10:40 ` Yunsheng Lin
2019-10-15 16:58 ` Greg KH
2019-10-16 12:07 ` Yunsheng Lin
2019-10-28 9:20 ` Yunsheng Lin
2019-10-29 8:53 ` Michal Hocko
2019-10-30 1:58 ` Yunsheng Lin
2019-10-10 8:56 ` Peter Zijlstra
2019-09-25 10:40 ` Peter Zijlstra
2019-09-25 13:25 ` Michal Hocko
2019-09-25 16:31 ` Peter Zijlstra
2019-09-25 21:45 ` Peter Zijlstra
2019-09-26 9:05 ` Peter Zijlstra
2019-09-26 12:10 ` Peter Zijlstra
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=8cff8350-311e-3817-0c42-b6f98de589a3@huawei.com \
--to=linyunsheng@huawei.com \
--cc=akpm@linux-foundation.org \
--cc=anshuman.khandual@arm.com \
--cc=axboe@kernel.dk \
--cc=benh@kernel.crashing.org \
--cc=borntraeger@de.ibm.com \
--cc=bp@alien8.de \
--cc=cai@lca.pw \
--cc=catalin.marinas@arm.com \
--cc=chenhc@lemote.com \
--cc=dalias@libc.org \
--cc=dave.hansen@linux.intel.com \
--cc=davem@davemloft.net \
--cc=dledford@redhat.com \
--cc=gor@linux.ibm.com \
--cc=gregkh@linuxfoundation.org \
--cc=heiko.carstens@de.ibm.com \
--cc=hpa@zytor.com \
--cc=ink@jurassic.park.msu.ru \
--cc=jeffrey.t.kirsher@intel.com \
--cc=jhogan@kernel.org \
--cc=jiaxun.yang@flygoat.com \
--cc=len.brown@intel.com \
--cc=linux-alpha@vger.kernel.org \
--cc=linux-arm-kernel@lists.infradead.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mips@vger.kernel.org \
--cc=linux-s390@vger.kernel.org \
--cc=linux-sh@vger.kernel.org \
--cc=linuxppc-dev@lists.ozlabs.org \
--cc=luto@kernel.org \
--cc=mattst88@gmail.com \
--cc=mhocko@kernel.org \
--cc=mingo@redhat.com \
--cc=mpe@ellerman.id.au \
--cc=mwb@linux.vnet.ibm.com \
--cc=naveen.n.rao@linux.vnet.ibm.com \
--cc=paul.burton@mips.com \
--cc=paulus@samba.org \
--cc=peterz@infradead.org \
--cc=rafael@kernel.org \
--cc=ralf@linux-mips.org \
--cc=robin.murphy@arm.com \
--cc=rppt@linux.ibm.com \
--cc=rth@twiddle.net \
--cc=sparclinux@vger.kernel.org \
--cc=tbogendoerfer@suse.de \
--cc=tglx@linutronix.de \
--cc=will@kernel.org \
--cc=x86@kernel.org \
--cc=ysato@users.sourceforge.jp \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).