Re: [RFC 0/3] Mirrored memory support for boot time allocations

Re: [RFC 0/3] Mirrored memory support for boot time allocations

[Tony Luck]

On Fri, Feb 6, 2015 at 1:54 PM, Tony Luck <tony.luck@intel.com> wrote:
> Platforms that support a mix of mirrored and regular memory are coming.

Obviously I don't do enough -mm work to remember where linux-mm mailing list
is hosted :-(

Let's see who finds this on the linux-kernel list (that I did spell
right).  When v2
happens I'll get it to the right places.

-Tony

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC 0/3] Mirrored memory support for boot time allocations

[Xishi Qiu]

Add linux-mm@kvack.org

On 2015/5/18 15:58, Xishi Qiu wrote:

> On 2015/2/7 5:54, Tony Luck wrote:
> 
>> Platforms that support a mix of mirrored and regular memory are coming.
>>
>> We'd like to use the mirrored memory for kernel code, data and dynamically
>> allocated data because our machine check recovery code cannot fix problems
>> there.  This series modifies the memblock allocator to comprehend mirrored
>> memory and use it for all boot time allocations.  Later I'll dig into page_alloc.c
>> to put the leftover mirrored memory into a zone to be used for kernel allocation
>> by slab/slob/slub and others.
> 
> Hi Tony,
> 
> Is it means that you will create a new zone to fill mirrored memory, like the
> movable zone, right? 
> I think this will change a lot of code, why not create a new migrate type?
> such as CMA, e.g. MIGRATE_MIRROR
> 
> Thanks,
> Xishi Qiu
> 
>>
>> You'll see why this is just RFC when you get to part 3.
>>
>> Tony Luck (3):
>>   mm/memblock: Add extra "flag" to memblock to allow selection of memory
>>     based on attribute
>>   mm/memblock: Allocate boot time data structures from mirrored memory
>>   x86, mirror: x86 enabling - find mirrored memory ranges and tell
>>     memblock
>>
>>  arch/s390/kernel/crash_dump.c |   4 +-
>>  arch/sparc/mm/init_64.c       |   4 +-
>>  arch/x86/kernel/check.c       |   2 +-
>>  arch/x86/kernel/e820.c        |   2 +-
>>  arch/x86/mm/init_32.c         |   2 +-
>>  arch/x86/mm/memtest.c         |   2 +-
>>  include/linux/memblock.h      |  43 ++++++++++------
>>  mm/cma.c                      |   4 +-
>>  mm/memblock.c                 | 113 ++++++++++++++++++++++++++++++++----------
>>  mm/nobootmem.c                |  12 ++++-
>>  10 files changed, 135 insertions(+), 53 deletions(-)
>>
> 
> 
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
> 
> .
> 



--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

RE: [RFC 3/3] x86, mirror: x86 enabling - find mirrored memory ranges and tell memblock

[Luck, Tony]

On 2015/2/4 6:40, Tony Luck wrote:
>> Can't post this part yet because it uses things in an upcoming[*] ACPI, UEFI, or some
>> other four-letter-ending-in-I standard.  So just imagine a call someplace early
>> in startup that reads information about mirrored address ranges and does:
>> 

> Does the upcoming[*] ACPI will add a new flag in SRAT tables? just like memory hotplug.
>
> #define ACPI_SRAT_MEM_HOT_PLUGGABLE (1<<1)	/* 01: Memory region is hot pluggable */
> +#define ACPI_SRAT_MEM_MIRROR	    (1<<3)	/* 03: Memory region is mirrored */

The choice for this was UEFI - new attribute bit in the GetMemoryMap() return value.

UEFI 2.5 has been published with this change and I posted a newer patch 10 days ago:

  https://lkml.org/lkml/2015/5/8/521

-Tony

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

RE: [RFC 0/3] Mirrored memory support for boot time allocations

[Luck, Tony]

> Is it means that you will create a new zone to fill mirrored memory, like the
> movable zone, right? 

That's my general plan.

> I think this will change a lot of code, why not create a new migrate type?
> such as CMA, e.g. MIGRATE_MIRROR

I'm still exploring options ... the idea is to use mirrored memory for kernel allocations
(because our machine check recovery code will always crash the system for errors
in kernel memory - while we can avoid the crash for errors in application memory).
I'm not familiar with CMA ... can you explain a bit how it might let me direct kernel
allocations to specific areas of memory?

-Tony

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>