linux-kernel - RE: [PATCH V2 4/6] mm: rename the global section array to mem

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: <TYYPR01MB67770ED42E65667E304F2FF9DD3D9@TYYPR01MB6777.jpnprd01.prod.outlook.com>
Date:   Wed, 2 Jun 2021 05:02:31 +0000
From:   HAGIO KAZUHITO(萩尾　一仁) 
        <k-hagio-ab@....com>
To:     Baoquan He <bhe@...hat.com>
CC:     Andrew Morton <akpm@...ux-foundation.org>,
        David Hildenbrand <david@...hat.com>,
        Dong Aisheng <dongas86@...il.com>,
        Dong Aisheng <aisheng.dong@....com>,
        "linux-mm@...ck.org" <linux-mm@...ck.org>,
        open list <linux-kernel@...r.kernel.org>,
        Dave Young <dyoung@...hat.com>,
        Vivek Goyal <vgoyal@...hat.com>,
        "kexec@...ts.infradead.org" <kexec@...ts.infradead.org>
Subject: RE: [PATCH V2 4/6] mm: rename the global section array to
 mem_sections

-----Original Message-----
> On 06/02/21 at 01:11am, HAGIO KAZUHITO(萩尾 一仁) wrote:
> > -----Original Message-----
> > > On Tue, 1 Jun 2021 10:40:09 +0200 David Hildenbrand <david@...hat.com> wrote:
> > >
> > > > > Thanks, i explained the reason during my last reply.
> > > > > Andrew has already picked this patch to -mm tree.
> > > >
> > > > Just because it's in Andrews tree doesn't mean it will end up upstream. ;)
> > > >
> > > > Anyhow, no really strong opinion, it's simply unnecessary code churn
> > > > that makes bisecting harder without real value IMHO.
> > >
> > > I think it's a good change - mem_sections refers to multiple instances
> > > of a mem_section.  Churn is a pain, but that's the price we pay for more
> > > readable code.  And for having screwed it up originally ;)
> >
> > From a makedumpfile/crash-utility viewpoint, I don't deny kernel improvement
> > and probably the change will not be hard for them to support, but I'd like
> > you to remember that the tool users will need to update them for the change.
> 
> As VIM user, I can understand Aisheng's feeling on the mem_section
> variable which has the same symbol name as its type. Meanwhile it does
> cause makedumpfile/crash having to be changed accordingly.
> 
> Maybe we can carry it when any essential change is needed in both kernel
> and makedumpfile/crash around it.

Yes, that is a possible option.

> 
> >
> > The situation where we need to update the tools for new kernels is usual, but
> > there are not many cases that they cannot even start session, and this change
> 
> By the way, Kazu, about a case starting session, could you be more specific
> or rephrase? I may not get it clearly. Thanks.

As for the current crash, the "mem_section" symbol is used to determine
which memory model is used.

        if (kernel_symbol_exists("mem_section"))
                vt->flags |= SPARSEMEM;
        else if (kernel_symbol_exists("mem_map")) {
                get_symbol_data("mem_map", sizeof(char *), &vt->mem_map);
                vt->flags |= FLATMEM;
        } else
                vt->flags |= DISCONTIGMEM;

So without updating, crash will assume that the memory model is DISCONTIGMEM,
fail during vm_init() and cannot start a session.  This is an imitation of
the situation though:

-       if (kernel_symbol_exists("mem_section"))
+       if (kernel_symbol_exists("mem_sectionX"))

# crash
...
crash: invalid structure member offset: pglist_data_node_mem_map
           FILE: memory.c  LINE: 16420  FUNCTION: dump_memory_nodes()

[/root/bin/crash] error trace: 465304 => 4ac2bf => 4aae19 => 57f4d7

  57f4d7: OFFSET_verify+164
  4aae19: dump_memory_nodes+5321
  4ac2bf: vm_init+4031
  465304: main_loop+392

#

Every time a kernel is released, there are some changes that crash can
start up with but cannot run a specific crash's command, but a change
that crash cannot start up like this case does not occur often.

Also as for makedumpfile, the "SYMBOL(mem_section)" vmcore entry is used
to determine the memory model, so it will fail with the following error
without an update.

# ./makedumpfile --mem-usage /proc/kcore 
get_mem_map: Can't distinguish the memory type.

makedumpfile Failed.

Thanks,
Kazu