[PATCH] x86: keep the crash kernel below 896 MiB

[PATCH] x86: keep the crash kernel below 896 MiB

[Warner Wang]

When use the crashkernel=size parameter without specifying a "@xM"
suffix, the crashkernel will be reserved at the highest suitable
address of memory, this will make the kexec-tools fail to work.
The current kexec-tools have a limitation that it can only work
with crashkernel reserved under 896 MiB.  we can increase this
limit once kexec-tools are fixed.
Linux kernel itself have this code for kexec-tools limitation,
so I think Xen code should also have it.

Signed-off-by: Warner Wang <warner.wang@hp.com>
---
 xen/arch/x86/setup.c |    7 +++++++
 1 files changed, 7 insertions(+), 0 deletions(-)

diff --git a/xen/arch/x86/setup.c b/xen/arch/x86/setup.c
index 43301a5..ae4c947 100644
--- a/xen/arch/x86/setup.c
+++ b/xen/arch/x86/setup.c
@@ -943,6 +943,13 @@ void __init __start_xen(unsigned long mbi_p)
             }
         }
 
+        /*
+         * Keep the crash kernel below this limit.  On 64 bits, kexec-tools
+         * currently limits us to 896 MiB; increase this limit once kexec-tools
+         * are fixed.
+         */
+        e = min_t(uint64_t, e, 896 << 20);
+
         /* Don't overlap with modules. */
         e = consider_modules(s, e, PAGE_ALIGN(kexec_crash_area.size),
                              mod, mbi->mods_count, -1);
-- 
1.7.1

Re: [PATCH] x86: keep the crash kernel below 896 MiB

[Jan Beulich]

>>> On 25.04.13 at 10:57, Warner Wang <warner.wang@hp.com> wrote:
> When use the crashkernel=size parameter without specifying a "@xM"
> suffix, the crashkernel will be reserved at the highest suitable
> address of memory, this will make the kexec-tools fail to work.
> The current kexec-tools have a limitation that it can only work
> with crashkernel reserved under 896 MiB.  we can increase this
> limit once kexec-tools are fixed.
> Linux kernel itself have this code for kexec-tools limitation,
> so I think Xen code should also have it.

Andrew,

didn't you post something along those lines at some point too,
yet there was a reason it never got applied?

Jan

> --- a/xen/arch/x86/setup.c
> +++ b/xen/arch/x86/setup.c
> @@ -943,6 +943,13 @@ void __init __start_xen(unsigned long mbi_p)
>              }
>          }
>  
> +        /*
> +         * Keep the crash kernel below this limit.  On 64 bits, kexec-tools
> +         * currently limits us to 896 MiB; increase this limit once kexec-tools
> +         * are fixed.
> +         */
> +        e = min_t(uint64_t, e, 896 << 20);
> +
>          /* Don't overlap with modules. */
>          e = consider_modules(s, e, PAGE_ALIGN(kexec_crash_area.size),
>                               mod, mbi->mods_count, -1);
> -- 
> 1.7.1

Re: [PATCH] x86: keep the crash kernel below 896 MiB

[Andrew Cooper]

On 25/04/13 10:51, Jan Beulich wrote:
>>>> On 25.04.13 at 10:57, Warner Wang <warner.wang@hp.com> wrote:
>> When use the crashkernel=size parameter without specifying a "@xM"
>> suffix, the crashkernel will be reserved at the highest suitable
>> address of memory, this will make the kexec-tools fail to work.
>> The current kexec-tools have a limitation that it can only work
>> with crashkernel reserved under 896 MiB.  we can increase this
>> limit once kexec-tools are fixed.
>> Linux kernel itself have this code for kexec-tools limitation,
>> so I think Xen code should also have it.
> Andrew,
>
> didn't you post something along those lines at some point too,
> yet there was a reason it never got applied?
>
> Jan

Jan:
I have some low crashinfo patches but they are in.  It was for
positioning of the Xen per-cpu crash notes and the console ring.

Warner:
I am curious to know what this magic 896MiB limit is.  Do you have a
reference to the problem.

~Andrew

>
>> --- a/xen/arch/x86/setup.c
>> +++ b/xen/arch/x86/setup.c
>> @@ -943,6 +943,13 @@ void __init __start_xen(unsigned long mbi_p)
>>              }
>>          }
>>  
>> +        /*
>> +         * Keep the crash kernel below this limit.  On 64 bits, kexec-tools
>> +         * currently limits us to 896 MiB; increase this limit once kexec-tools
>> +         * are fixed.
>> +         */
>> +        e = min_t(uint64_t, e, 896 << 20);
>> +
>>          /* Don't overlap with modules. */
>>          e = consider_modules(s, e, PAGE_ALIGN(kexec_crash_area.size),
>>                               mod, mbi->mods_count, -1);
>> -- 
>> 1.7.1
>
>

Re: [PATCH] x86: keep the crash kernel below 896 MiB

[David Vrabel]

On 25/04/13 09:57, Warner Wang wrote:
> When use the crashkernel=size parameter without specifying a "@xM"
> suffix, the crashkernel will be reserved at the highest suitable
> address of memory, this will make the kexec-tools fail to work.
> The current kexec-tools have a limitation that it can only work
> with crashkernel reserved under 896 MiB.  we can increase this
> limit once kexec-tools are fixed.
> Linux kernel itself have this code for kexec-tools limitation,
> so I think Xen code should also have it.

I don't think we should limit Xen in this way for any particular tool.

It's trivial to specify a location in the crashkernel parameter to work
around this particular tool limitation.

I also think that this limitation may well be fixed as of kexec-tools
2.0.4.  64-bit kernels are now loaded above 4 GiB.

David

> Signed-off-by: Warner Wang <warner.wang@hp.com>
> ---
>  xen/arch/x86/setup.c |    7 +++++++
>  1 files changed, 7 insertions(+), 0 deletions(-)
> 
> diff --git a/xen/arch/x86/setup.c b/xen/arch/x86/setup.c
> index 43301a5..ae4c947 100644
> --- a/xen/arch/x86/setup.c
> +++ b/xen/arch/x86/setup.c
> @@ -943,6 +943,13 @@ void __init __start_xen(unsigned long mbi_p)
>              }
>          }
>  
> +        /*
> +         * Keep the crash kernel below this limit.  On 64 bits, kexec-tools
> +         * currently limits us to 896 MiB; increase this limit once kexec-tools
> +         * are fixed.
> +         */
> +        e = min_t(uint64_t, e, 896 << 20);
> +
>          /* Don't overlap with modules. */
>          e = consider_modules(s, e, PAGE_ALIGN(kexec_crash_area.size),
>                               mod, mbi->mods_count, -1);