Re: [Qemu-devel] QEMU NVDIMM as type 7 in e820 table

[Haozhong Zhang]

On 07/28/17 13:45 -0600, Ross Zwisler wrote:
> On Fri, Jul 28, 2017 at 11:11:10AM -0700, Dan Williams wrote:
> > On Fri, Jul 28, 2017 at 11:04 AM, Ross Zwisler
> > <address@hidden> wrote:
> > > I've been using the virtualized NVDIMM support in QEMU for testing, and I
> > > noticed that the physical addresses used by the virtual NVDIMMs aren't 
> > > present
> > > in the guest's e820 table.
> > >
> > > Here is the e820 table on my QEMU instance where I have one 32 GiB virtual
> > > NVDIMM:
> > >
> > > [    0.000000] e820: BIOS-provided physical RAM map:
> > > [    0.000000] BIOS-e820: [mem 0x0000000000000000-0x000000000009fbff] 
> > > usable
> > > [    0.000000] BIOS-e820: [mem 0x000000000009fc00-0x000000000009ffff] 
> > > reserved
> > > [    0.000000] BIOS-e820: [mem 0x00000000000f0000-0x00000000000fffff] 
> > > reserved
> > > [    0.000000] BIOS-e820: [mem 0x0000000000100000-0x00000000bffdefff] 
> > > usable
> > > [    0.000000] BIOS-e820: [mem 0x00000000bffdf000-0x00000000bfffffff] 
> > > reserved
> > > [    0.000000] BIOS-e820: [mem 0x00000000feffc000-0x00000000feffffff] 
> > > reserved
> > > [    0.000000] BIOS-e820: [mem 0x00000000fffc0000-0x00000000ffffffff] 
> > > reserved
> > > [    0.000000] BIOS-e820: [mem 0x0000000100000000-0x000000023fffffff] 
> > > usable
> > >
> > > The physical addresses used by the virtual NVDIMM are 
> > > 0x240000000-0xA40000000.
> > > You can see this by looking at ndctl and the values we get from the NFIT:
> > >
> > > # ndctl list -R
> > > {
> > >   "dev":"region0",
> > >   "size":34359738368,
> > >   "available_size":0,
> > >   "type":"pmem"
> > > }
> > >
> > > # grep . /sys/bus/nd/devices/region0/{resource,size}
> > > region0/resource:0x240000000
> > > region0/size:34359738368
> > >
> > > Or you can see the same info by using iasl to dump
> > > /sys/firmware/acpi/tables/NFIT:
> > >
> > > [028h 0040   2]                Subtable Type : 0000 [System Physical 
> > > Address Range]
> > > [02Ah 0042   2]                       Length : 0038
> > >
> > > [02Ch 0044   2]                  Range Index : 0002
> > > [02Eh 0046   2]        Flags (decoded below) : 0003
> > >                    Add/Online Operation Only : 1
> > >                       Proximity Domain Valid : 1
> > > [030h 0048   4]                     Reserved : 00000000
> > > [034h 0052   4]             Proximity Domain : 00000000
> > > [038h 0056  16]           Address Range GUID : 
> > > 66F0D379-B4F3-4074-AC43-0D3318B78CDB
> > > [048h 0072   8]           Address Range Base : 0000000240000000
> > > [050h 0080   8]         Address Range Length : 0000000800000000
> > > [058h 0088   8]         Memory Map Attribute : 0000000000008008
> > >
> > > I expected to see a type 7 region for the NVDIMM physical address range 
> > > in the
> > > e820 table, so something like:
> > >
> > > [    0.000000] e820: BIOS-provided physical RAM map:
> > > [    0.000000] BIOS-e820: [mem 0x0000000000000000-0x000000000009fbff] 
> > > usable
> > > [    0.000000] BIOS-e820: [mem 0x000000000009fc00-0x000000000009ffff] 
> > > reserved
> > > [    0.000000] BIOS-e820: [mem 0x00000000000f0000-0x00000000000fffff] 
> > > reserved
> > > [    0.000000] BIOS-e820: [mem 0x0000000000100000-0x00000000bffdefff] 
> > > usable
> > > [    0.000000] BIOS-e820: [mem 0x00000000bffdf000-0x00000000bfffffff] 
> > > reserved
> > > [    0.000000] BIOS-e820: [mem 0x00000000feffc000-0x00000000feffffff] 
> > > reserved
> > > [    0.000000] BIOS-e820: [mem 0x00000000fffc0000-0x00000000ffffffff] 
> > > reserved
> > > [    0.000000] BIOS-e820: [mem 0x0000000100000000-0x000000023fffffff] 
> > > usable
> > > [    0.000000] BIOS-e820: [mem 0x0000000240000000-0x0000000A40000000] 
> > > persistent (type 7)
> > >
> > 
> > Do you need that informationin e820? Linux effectively ignores type-7.
> > As long as the range is treated as reserved it's not clear that you
> > need the e820 entry. We also infect the persistent type back into the
> > memory map when the NFIT driver loads. /proc/iomem should show the
> > right data.
> 
> [ Adding Linda & Toshi to see if they have an opinion. ]
> 
> I guess maybe we don't need it.  Yep, /proc/iomem looks good:
> 
>   # cat /proc/iomem
>   00000000-00000fff : Reserved
>   00001000-0009fbff : System RAM
>   ...
>   100000000-23fffffff : System RAM
>   240000000-a3fffffff : Persistent Memory
>     240000000-a3fffffff : namespace0.0
> 
> I was just worried that this was an inconsistency between the way that virtual
> NVDIMMs are presented vs the way that they will be presented on bare metal.  I
> at least look at the e820 table to get my bearings of how memory is laid out -
> maybe I just need to look at /proc/iomem instead?

Do any OS or applications rely on the E820 information or the
consistency between E820 and NFIT to properly work? If any, I can make
a QEMU patch to build type-7 e820 entries.

Haozhong