On 02/22/2015 06:27 PM, Christoph Hellwig wrote:
On Thu, Feb 19, 2015 at 11:25:37AM +0200, Boaz Harrosh wrote:
> I do not see why you need the nvdimm_type= kernel option at all.
> I have here a script that auto detects any NvDIMM. It works with all
> the chips that I have access to. And Also it has support for if you have
> For all these detected regions it will load a pmem device.
> It is easy to filter for any type of memory you want. What
> will the (annoying) kernel option give you?
> OK I might be jumping the guns, send the patch and I'll look
> at it.
The kernel option means we can autodetect the nvdimms in kernel space
with just that option set, no need for needing userspace setup.
Do you mean that pmem is:
- Loaded without any parameters.
- A new API is defined for enumerating NvDIMMs. pmem uses that for
creating new devices.
- e820 registers such devices according to special type passed on Kernel
command-line. (That could be a Kconfig as well you know, I hate Kernel
(Similar to what very old prd.c did only not hacking the e820 lists directly)
How does your script / patch work?
Easy just parses /proc/iomem + looks at Kernel command line.
It has support not only for type-12 memory but also for "reserved"
regions, made by memmap=sss$aaa stated on command line. And/or also
an /etc/pmem.cfg list of regions. So you can go automatic or manual or
a mix of both (And slice a region for xfstests scratch device).
In any case it helps to have my patches, but also old Kernels are
supported with the BLK_DEV_PMEM_IGNORE_REQUEST_MEM_RET. On old Kernels
they are all "reserved" regions, no unknown-12 type.
Once a list of regions or split-regions is established pmem is loaded
with that list.
I won't be back to my nvdimm enabled hardware until after
so any work from me in this area will have to wait a bit.
Have a good time. Will you guys have a talk about pmem ? If yes I
would love it if you guys can talk about the use-of-pages-with-pmem.
Please tell me I can write a looong argument about it.