Re: [PATCHv7] x86/kdump: bugfix, make the behavior of crashkernel=X consistent with kaslr

[Jerry Hoemann]

On Fri, Feb 01, 2019 at 12:47:40AM +0100, Borislav Petkov wrote:
> On Thu, Jan 31, 2019 at 03:27:32PM -0700, Jerry Hoemann wrote:
> > So even if a system administrator is diligent and tests
> > that a chosen kdump configuration works, that configuration
> > might not work on some random reboot 7 months in the future.
> 
> Jerry, did you read the rest of the thread where I'm *actually*
> suggesting to make the allocation code more robust against such
> failures?


Boris,

I may have misunderstood your earlier comment:

  So we don't really need this - we simply need to tell people to use high
  if it fails with KASLR, AFAICT

To imply an iterative approach to crashkernel size discovery.  Whereas you
may simply have ment:  Always use ,high as the old way is broken.


> Now let's look at the code:
> 
> The "high" allocation does:
> 
>                 crash_base = memblock_find_in_range(CRASH_ALIGN,
>                                                     high ? CRASH_ADDR_HIGH_MAX
>                                                          : CRASH_ADDR_LOW_MAX,
>                                                     crash_size, CRASH_ALIGN);
> 
> where high=true and CRASH_ADDR_HIGH_MAX on 64-bit is MAXMEM:
> 
> # define CRASH_ADDR_HIGH_MAX    MAXMEM
> 
> The second fallback in the suggested patch does the same:
> 
> +               /*
> +                * crashkernel=X reserve below 4G fails? Try MAXMEM
> +                */
> +               if (!high && !crash_base)
> +                       crash_base = memblock_find_in_range(CRASH_ALIGN,
> +                                               CRASH_ADDR_HIGH_MAX,
> +                                               crash_size, CRASH_ALIGN);
> 
> and yet I get back that falling back to "high" if the first allocation
> doesn't succeed is not something we should do by default because of
> reasons. But this patch *practically* *does* it.


Is your objection only to the second fallback of allocating
memory above >= 4GB?   Or are you objecting to allocating from
(896 .. 4GB) as well?

Falling back to allocating < 4GB probably satisfes most of the cases
where the original allocation fails.

thanks

-- 

-----------------------------------------------------------------------------
Jerry Hoemann                  Software Engineer   Hewlett Packard Enterprise
-----------------------------------------------------------------------------