On Tue, Apr 28, 2015 at 3:58 PM, Andy Lutomirski <luto(a)amacapital.net> wrote:
On Tue, Apr 28, 2015 at 3:21 PM, Phil Pokorny
> On Tue, Apr 28, 2015 at 2:04 PM, Andy Lutomirski <luto(a)amacapital.net> wrote:
>> On Tue, Apr 28, 2015 at 11:25 AM, Dan Williams <dan.j.williams(a)intel.com>
>>> +config ND_E820
>>> + tristate "E820: Support the E820-type-12 PMEM convention"
>>> + depends on X86_PMEM_LEGACY
>>> + default m if X86_PMEM_LEGACY
>>> + select LIBND
>>> + help
>>> + Prior to ACPI 6 some platforms advertised peristent memory
>>> + via type-12 e820 memory ranges. Create a libnd bus and
>>> + attach an instance of the pmem driver to these ranges.
>> How about something like:
>> "This driver allows libnd to work with legacy, pre-ACPI 6 NVDIMMs.
>> This enables such devices to be exposed as block devices using PMEM.
>> The legacy NVDIMM interface is problematic. This driver will not work
>> if you boot using UEFI, and some NVDIMMs and motherboards that work
>> with this driver may require proprietary code in order to work
> Perhaps not "problematic" but "requires a BIOS in Legacy mode"
> It might also mention that if you use the kernel command line
> memmap=nn!ss syntax it adds
> a type 12 region to the e820 map and so you would want this support.
> If you have a motherboard with UEFI support for NVDIMM's that would be
> the recommended
This is such a mess that I think this driver should maybe flat-out
refuse to load in this type of configuration without some scary module
option. I have some NVDIMMs that report as type 12 but need two extra
out-of-tree drivers to work safely. First, they need i2c_imc or the
equivalent (I'll try to resubmit that soon). Second, they need secret
magic NDAed register poking. The latter is very problematic.
My current experience is that things may be changing to something of a de-facto
standard in the area of register poking. In which case, we should be
able to ask
the de-facto vendor standard to be released under a non-NDA license so we can
write a proper user-space library for it. Or at worst, get a
proprietary source utility
that can do the poking.
The vendor isn't going to sell anything if they don't provide the
tools their resellers
and customers need.
At the very least, I think we should discourage people who don't
really know what they're doing from using this driver without care.
What would be the fun in that...
But seriously, speaking as Penguin Computing and a retailer of
hardware, I'd rather
not have the kernel telling my customers what's safe and what isn't
when it's a matter
of opinion. We provide a solution with support and having to tell my
need to load the module with the 'THIS_IS_UNSAFE' argument set to 3"
Another intersesting possibility of the memmap= directive to declare a
type 12 region of
of memory is that you can test the driver (without the persistance) on
any arbitrary region
of memory in a machine. Other comments on this patch set talked about
having to put
virtual test hardware in qemu or kvm. Aside from the register poking,
memmap=xx!yy to the command line gives you something pmem can attach to and
you can use to test with. I suppose you could even simulate
persistance by saving off
the contents and restoring it on a controlled reboot.
Philip Pokorny, RHCE
Chief Technology Officer
PENGUIN COMPUTING, Inc