Re: [PATCH 0/4] Adding support to parse BERT for libnvdimm

["Kani, Toshi" <toshi.kani-ZPxbGqLxI0U@public.gmane.org>]

On Fri, 2018-03-30 at 09:49 -0700, Dave Jiang wrote:
> 
> On 03/30/2018 09:45 AM, Kani, Toshi wrote:
> > On Fri, 2018-03-30 at 09:38 -0700, Dan Williams wrote:
> > > On Fri, Mar 30, 2018 at 8:04 AM, Kani, Toshi <toshi.kani-ZPxbGqLxI0U@public.gmane.org> wrote:
> > > > On Thu, 2018-03-29 at 15:37 -0700, Dave Jiang wrote:
> > > > > The following series implements support for parsing of the BERT records
> > > > > and adding the error memory ranges to nvdimm badblocks in order for the
> > > > > kernel to avoid prevent the kernel from accessing those areas. And with
> > > > > the addition of this support, we can surface the nd regions instead of waiting
> > > > > for ARS to complete. So the ARS handling is reworked to run in the
> > > > > background and not block nd region registration.
> > > > 
> > > > Hi Dave,
> > > > 
> > > > I agree on the problem, and adding an ability to obtain pmem badblocks
> > > > records at boot-time without waiting for a new ARS scan to complete is a
> > > > good option for users.
> > > > 
> > > > However, I do not think using the BERT table is a good approach.  This
> > > > requires FW to report pmem badblocks records with a new interface in
> > > > addition to ARS records, which FW already implements for pmem.  ACPI 6.2
> > > > defines Start ARS with Flags Bit[1] set to report badblocks record
> > > > without starting a new ARS scan.  We set this bit after receiving a 0x81
> > > > notification at this point.
> > > > 
> > > > Can we use ARS with Flags bit[1] set at boot-time so that both OS and FW
> > > > can use the same ARS implementation?
> > > 
> > > You have a point.
> > > 
> > > The other benefit I see to this policy is that it hopefully convinces
> > > BIOS implementations to not run ARS at boot and leave it to the OS to
> > > manage it in the background. If the platform has any critical errors
> > > to report, i.e. ones that triggered a system reset, then it should be
> > > able to report them in the flag-bit1 case.
> > 
> > Agreed.
> > 
> > > This also lets the implementation be completely self contained to the
> > > nfit driver, and not grow any BERT entanglements that may or may not
> > > be valid for the persistent memory case.
> > 
> > Right.  Thanks Dan!
> > -Toshi
> > 
> 
> Ok, I'll respin the series with what you suggested Toshi and drop the
> BERT stuff.

Great!  Thanks Dave!
-Toshi