VMA/Paging apis broken on ARM/ARM64, possible work-arounds?

VMA/Paging apis broken on ARM/ARM64, possible work-arounds?

[Kamil Leoniak]

Hi,

I have this problem with the kernel I'm working with. Normally PTEs
are made for every VM addr (ktext, modules and so on). On ARM/ARM64
certain VM addrs (in kernel) are allocated only with PMDs, they not
have a corresponding PTE cells.

This breaks a whole range of kernel APIs, for example "set_memory_rw".
It will work for any region that has a PTE, but if it's a direct PMD
mapping, it will fail.

Is there some simple work-around for that? Can I somehow remap a
specific VM address into a specific physical address? (to override the
"simple" PMD mapping).

Thank you!

VMA/Paging apis broken on ARM/ARM64, possible work-arounds?

[Russell King - ARM Linux]

On Tue, Jun 05, 2018 at 02:40:28PM +0200, Kamil Leoniak wrote:
> I have this problem with the kernel I'm working with. Normally PTEs
> are made for every VM addr (ktext, modules and so on). On ARM/ARM64
> certain VM addrs (in kernel) are allocated only with PMDs, they not
> have a corresponding PTE cells.
> 
> This breaks a whole range of kernel APIs, for example "set_memory_rw".

These return -EINVAL if you try to use them on a region that isn't
supported by the implementation.  Their use on the kernel is not
supported, sorry.

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 8.8Mbps down 630kbps up
According to speedtest.net: 8.21Mbps down 510kbps up

VMA/Paging apis broken on ARM/ARM64, possible work-arounds?

[Mark Rutland]

On Tue, Jun 05, 2018 at 02:40:28PM +0200, Kamil Leoniak wrote:
> Hi,

Hi,

> I have this problem with the kernel I'm working with. Normally PTEs
> are made for every VM addr (ktext, modules and so on). On ARM/ARM64
> certain VM addrs (in kernel) are allocated only with PMDs, they not
> have a corresponding PTE cells.
> 
> This breaks a whole range of kernel APIs, for example "set_memory_rw".
> It will work for any region that has a PTE, but if it's a direct PMD
> mapping, it will fail.

As Russell said, for various reasons, APIs which change the attributes
or permissions of pages aren't supported on arbitrary mappings.

> Is there some simple work-around for that? Can I somehow remap a
> specific VM address into a specific physical address? (to override the
> "simple" PMD mapping).

Could you please explain your use-case?

Why are you trying to remap a range that already has a mapping?

Thanks,
Mark.