linux-kernel - Re: [PATCH RFC v1] mm: is_mem_section

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 for Android: free password hash cracker in your pocket

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

Message-ID: <65606e2e-1cf7-de3b-10b1-33653cb41a52@redhat.com>
Date:   Fri, 17 Jan 2020 15:58:26 +0100
From:   David Hildenbrand <david@...hat.com>
To:     Michal Hocko <mhocko@...nel.org>
Cc:     linux-kernel@...r.kernel.org, linux-mm@...ck.org,
        Benjamin Herrenschmidt <benh@...nel.crashing.org>,
        Paul Mackerras <paulus@...ba.org>,
        Michael Ellerman <mpe@...erman.id.au>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        "Rafael J. Wysocki" <rafael@...nel.org>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Leonardo Bras <leonardo@...ux.ibm.com>,
        Nathan Lynch <nathanl@...ux.ibm.com>,
        Allison Randal <allison@...utok.net>,
        Nathan Fontenot <nfont@...ux.vnet.ibm.com>,
        Thomas Gleixner <tglx@...utronix.de>,
        Dan Williams <dan.j.williams@...el.com>,
        Stephen Rothwell <sfr@...b.auug.org.au>,
        Anshuman Khandual <anshuman.khandual@....com>,
        lantianyu1986@...il.com, linuxppc-dev@...ts.ozlabs.org
Subject: Re: [PATCH RFC v1] mm: is_mem_section_removable() overhaul

On 17.01.20 15:52, Michal Hocko wrote:
> On Fri 17-01-20 14:08:06, David Hildenbrand wrote:
>> On 17.01.20 12:33, Michal Hocko wrote:
>>> On Fri 17-01-20 11:57:59, David Hildenbrand wrote:
>>>> Let's refactor that code. We want to check if we can offline memory
>>>> blocks. Add a new function is_mem_section_offlineable() for that and
>>>> make it call is_mem_section_offlineable() for each contained section.
>>>> Within is_mem_section_offlineable(), add some more sanity checks and
>>>> directly bail out if the section contains holes or if it spans multiple
>>>> zones.
>>>
>>> I didn't read the patch (yet) but I am wondering. If we want to touch
>>> this code, can we simply always return true there? I mean whoever
>>> depends on this check is racy and the failure can happen even after
>>> the sysfs says good to go, right? The check is essentially as expensive
>>> as calling the offlining code itself. So the only usecase I can think of
>>> is a dumb driver to crawl over blocks and check which is removable and
>>> try to hotremove it. But just trying to offline one block after another
>>> is essentially going to achieve the same.
>>
>> Some thoughts:
>>
>> 1. It allows you to check if memory is likely to be offlineable without
>> doing expensive locking and trying to isolate pages (meaning:
>> zone->lock, mem_hotplug_lock. but also, calling drain_all_pages()
>> when isolating)
>>
>> 2. There are use cases that want to identify a memory block/DIMM to
>> unplug. One example is PPC DLPAR code (see this patch). Going over all
>> memory block trying to offline them is an expensive operation.
>>
>> 3. powerpc-utils (https://github.com/ibm-power-utilities/powerpc-utils)
>> makes use of /sys/.../removable to speed up the search AFAIK.
> 
> Well, while I do see those points I am not really sure they are worth
> having a broken (by-definition) interface.

It's a pure speedup. And for that, the interface has been working
perfectly fine for years?

>  
>> 4. lsmem displays/groups by "removable".
> 
> Is anybody really using that?

Well at least I am using that when testing to identify which
(ZONE_NORMAL!) block I can easily offline/re-online (e.g., to validate
all the zone shrinking stuff I have been fixing)

So there is at least one user ;)

[...]
> 
>>> Or does anybody see any reasonable usecase that would break if we did
>>> that unconditional behavior?
>>
>> If we would return always "true", then the whole reason the
>> interface originally was introduced would be "broken" (meaning, less
>> performant as you would try to offline any memory block).
> 
> I would argue that the whole interface is broken ;). Not the first time
> in the kernel development history and not the last time either. What I
> am trying to say here is that unless there are _real_ usecases depending
> on knowing that something surely is _not_ offlineable then I would just
> try to drop the functionality while preserving the interface and see
> what happens.

I can see that, but I can perfectly well understand why - especially
powerpc - wants a fast way to sense which blocks actually sense to try
to online.

The original patch correctly states
   "which sections of
    memory are likely to be removable before attempting the potentially
    expensive operation."

It works as designed I would say.

-- 
Thanks,

David / dhildenb