Re: [PATCH] KVM: arm/arm64: Properly check for MMIO regions

[James Morse]

Hi KarimAllah,

On 12/07/2019 09:22, KarimAllah Ahmed wrote:
> Valid RAM can live outside kernel control (e.g. using "mem=" command-line
> parameter). This memory can still be used as valid guest memory for KVM. So
> ensure that we validate that this memory is definitely not "RAM" before
> assuming that it is an MMIO region.
> 
> One way to use memory outside kernel control is:
> 
> 1- Pass 'mem=' in the kernel command-line to limit the amount of memory managed
>    by the kernel.

"mem=" is a debug option, we ignore it if we need something located outside the 'mem=' region.


> 2- Map this physical memory you want to give to the guest with:
>    mmap("/dev/mem", physical_address_offset, ..)

/dev/mem is an egregious hack! If you need to use it, you probably didn't want an
operating-system in the first place.


> 3- Use the user-space virtual address as the "userspace_addr" field in
>    KVM_SET_USER_MEMORY_REGION ioctl.


... What do you want to do this for?

At a guess: this is to save all that annoying 'memory allocation' overhead at guest
startup. If you get your VMM to use hugetlbfs, you can reserve the memory during boot. I
do this with "hugepagesz=2M hugepages=512" on the kernel command-line.

(if you get a RAS error affecting memory that the kernel doesn't know about, it will
ignore it. Using hugetlbfs instead gives you all the good things: hugepage-splitting,
signals to your VMM, stage2 unmapping etc.)


Thanks,

James