Date:   Mon, 10 Jun 2019 14:51:45 +0800
From:   Dave Young <dyoung@...hat.com>
To:     Borislav Petkov <bp@...en8.de>
Cc:     Pingfan Liu <kernelfans@...il.com>, kexec@...ts.infradead.org,
        Baoquan He <bhe@...hat.com>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Mike Rapoport <rppt@...ux.vnet.ibm.com>, yinghai@...nel.org,
        vgoyal@...hat.com, Randy Dunlap <rdunlap@...radead.org>,
        x86@...nel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCHv7] x86/kdump: bugfix, make the behavior of crashkernel=X
 consistent with kaslr

On 06/07/19 at 07:30pm, Borislav Petkov wrote:
> On Mon, Jan 28, 2019 at 11:18:31AM +0100, Borislav Petkov wrote:
> > On Mon, Jan 28, 2019 at 05:58:09PM +0800, Dave Young wrote:
> > > Another reason is in case ,high we will need automatically reserve a
> > > region in low area for swiotlb.  So for example one use
> > > crashkernel=256M,high,  actual reserved memory is 256M above 4G and
> > > another 256M under 4G for swiotlb.  Normally it is not necessary for
> > > most people.  Thus we can not make ,high as default.
> > 
> > And how is the poor user to figure out that we decided for her/him that
> > swiotlb reservation is something not necessary for most people and thus
> > we fail the crashkernel= reservation?
> > 
> > IOW, that "logic" above doesn't make a whole lot of sense to me from
> > user friendliness perspective.
> 
> So to show what I mean: I'm trying to reserve a crash kernel region on a
> box here. I tried:
> 
> crashkernel=64M@16M
> 
> as it is stated in Documentation/kdump/kdump.txt.
> 
> Box said:
> 
> [    0.000000] crashkernel reservation failed - memory is in use.
> 
> Oh great.
> 
> Then I tried:
> 
> crashkernel=64M@64M
> 
> Box said:
> 
> [    0.000000] crashkernel reservation failed - memory is in use.
> 
> So I simply did:
> 
> crashkernel=64M
> 
> and the box said:
> 
> [    0.000000] Reserving 64MB of memory at 3392MB for crashkernel (System RAM: 16271MB)
> 
> So I could've gone a long time poking at the memory to find a suitable
> address.
> 
> So do you see what I mean with making this as user-friendly and as
> robust as possible?

Yes, it is clear to me, I absolutely agree that is not friendly :)

Previously without KASLR, one can check /proc/iomem to find a possible
free area and use it for next and future boot.  But in case KASLR
enabled nowadays it become harder to predict the persistent free areas.

> 
> In this case I don't care about *where* my crash kernel is - I only want
> to have one loaded *somewhere*.

We would suggest people to use crashkernel=X instead.  for the X@Y
I believe it is some historic thing, it *should* be able to be obsolete
at least on X86, (not sure other arches).
I expect people can comment if they have some use cases requiring this
X@Y way. 

We have modified the crashkernel=X to search 0 - 4G memory instead
of old 0 - 896M for low memory areas, so a possible case is people who
uses very old kexec-tools which can only load kernel to memory under
896M.

Another way is we just obsolete X@Y, but introduce another interface
like crahskernel=X,max=  (max will be used like the CRASH_ADDR_HIGH_MAX
in arch/x86/kernel/setup.c)

> 
> And the same strategy should be applied to other reservation attempts
> - we should try hard to reserve and if we cannot reserve, then try an
> alternating range.
> 
> I even think that
> 
> crashkernel=X@Y
> 
> should not simply fail if Y is occupied but keep trying and say
> 
> [    0.000000] Reserving 64MB of memory at alternative address 3392MB for crashkernel (System RAM: 16271MB)
> 
> and only fail when the user doesn't really want the kernel to try hard
> by booting with
> 
> crashkernel=X@Y,strict
> 
> But that's for another day.

Maybe X@Y,max=..  Then kernel will search begin with Y, and stop until
max - 1;

> 
> -- 
> Regards/Gruss,
>     Boris.
> 
> Good mailing practices for 400: avoid top-posting and trim the reply.

Thanks
Dave

Powered by blists - more mailing lists