linux-kernel - Re: [PATCH 4/4] dax: "Hotplug" persistent memory for use like normal RAM

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] [thread-next>] [day] [month] [year] [list]

Message-ID: <74c01a55-1a47-d649-a32a-ea597c24ab4a@linux.intel.com>
Date:   Fri, 18 Jan 2019 15:47:32 +0800
From:   Yanmin Zhang <yanmin_zhang@...ux.intel.com>
To:     Dave Hansen <dave.hansen@...el.com>,
        Dave Hansen <dave.hansen@...ux.intel.com>, dave@...1.net
Cc:     dan.j.williams@...el.com, dave.jiang@...el.com, zwisler@...nel.org,
        vishal.l.verma@...el.com, thomas.lendacky@....com,
        akpm@...ux-foundation.org, mhocko@...e.com,
        linux-nvdimm@...ts.01.org, linux-kernel@...r.kernel.org,
        linux-mm@...ck.org, ying.huang@...el.com, fengguang.wu@...el.com,
        bp@...e.de, bhelgaas@...gle.com, baiyaowei@...s.chinamobile.com,
        tiwai@...e.de
Subject: Re: [PATCH 4/4] dax: "Hotplug" persistent memory for use like normal
 RAM

On 2019/1/17 下午11:17, Dave Hansen wrote:
> On 1/17/19 12:19 AM, Yanmin Zhang wrote:
>>>
>> I didn't try pmem and I am wondering it's slower than DRAM.
>> Should a flag, such like _GFP_PMEM, be added to distinguish it from
>> DRAM?
> 
> Absolutely not. :)
Agree.

> 
> We already have performance-differentiated memory, and lots of ways to
> enumerate and select it in the kernel (all of our NUMA infrastructure).
Kernel does manage memory like what you say.
My question is: with your patch, PMEM becomes normal RAM, then there is
a chance for kernel to allocate PMEM as DMA buffer.
Some super speed devices like 10Giga NIC, USB (SSIC connecting modem), 
might not work well if DMA buffer is in PMEM as it's slower than DRAM.

Should your patchset consider it?



> 
> PMEM is also just the first of many "kinds" of memory that folks want to
> build in systems and use a "RAM".  We literally don't have space to put
> a flag in for each type.
> 
>