RE: nvdimm,pmem: makedumpfile: __vtop4_x86_64: Can't get a valid pte.

[Dan Williams <dan.j.williams@intel.com>]

lizhijian@fujitsu.com wrote:
> Hi folks,
> 
> I'm going to make crash coredump support pmem region. So
> I have modified kexec-tools to add pmem region to PT_LOAD of vmcore.
> 
> But it failed at makedumpfile, log are as following:
> 
> In my environment, i found the last 512 pages in pmem region will cause the error.
> 
> qemu commandline:
>   -object memory-backend-file,id=memnvdimm0,prealloc=yes,mem-path=/root/qemu-dax.img,share=yes,size=4267704320,align=2097152
> -device nvdimm,node=0,label-size=4194304,memdev=memnvdimm0,id=nvdimm0,slot=0
> 
> ndctl info:
> [root@rdma-server ~]# ndctl list
> [
>    {
>      "dev":"namespace0.0",
>      "mode":"devdax",
>      "map":"dev",
>      "size":4127195136,
>      "uuid":"f6fc1e86-ac5b-48d8-9cda-4888a33158f9",
>      "chardev":"dax0.0",
>      "align":4096
>    }
> ]
> [root@rdma-server ~]# ndctl list -iRD
> {
>    "dimms":[
>      {
>        "dev":"nmem0",
>        "id":"8680-56341200",
>        "handle":1,
>        "phys_id":0
>      }
>    ],
>    "regions":[
>      {
>        "dev":"region0",
>        "size":4263510016,
>        "align":16777216,
>        "available_size":0,
>        "max_available_extent":0,
>        "type":"pmem",
>        "iset_id":10248187106440278,
>        "mappings":[
>          {
>            "dimm":"nmem0",
>            "offset":0,
>            "length":4263510016,
>            "position":0
>          }
>        ],
>        "persistence_domain":"unknown"
>      }
>    ]
> }
> 
> iomem info:
> [root@rdma-server ~]# cat /proc/iomem  | grep Persi
> 140000000-23e1fffff : Persistent Memory
> 
> makedumpfile info:
> [   57.229110] kdump.sh[240]: mem_map[  71] ffffea0008e00000           238000           23e200
> 
> 
> Firstly, i wonder that
> 1) makedumpfile read the whole range of iomem(same with the PT_LOAD of pmem)
> 2) 1st kernel side only setup mem_map(vmemmap) for this namespace, which size is 512 pages smaller than iomem for some reasons.
> 3) Since there is an align in nvdimm region(16MiB in above), i also guess the maximum size of the pmem can used by user should
> be ALIGN(iomem, 10MiB), after this alignment, the last 512 pages will be dropped. then kernel only setups vmemmap for this
> range. but i didn't see any code doing such things in kernel side.
> 
> So if you guy know the reasons, please let me know :), any hint/feedback is very welcome.

This is due to the region alignment.

2522afb86a8c libnvdimm/region: Introduce an 'align' attribute

If you want to use the full capacity it would be something like this
(untested, and may destroy any data currently on the namespace):

ndctl destroy-namespace namespace0.0
echo $((2<<20)) > /sys/bus/nd/devices/region0/align
ndctl create-namespace -m dax -a 4k -M mem