On 2025/3/20 上午11:30, bibo mao wrote:
On 2025/3/19 上午3:40, Oliver Upton wrote:
On Tue, Mar 18, 2025 at 08:35:38PM +0100, David Hildenbrand wrote:
On 18.03.25 20:27, Catalin Marinas wrote:
On Tue, Mar 18, 2025 at 09:55:27AM -0300, Jason Gunthorpe wrote:
On Tue, Mar 18, 2025 at 09:39:30AM +0000, Marc Zyngier wrote:
The memslot must also be created with a new flag ((2c) in the
taxonomy
above) that carries the "Please map VM_PFNMAP VMAs as cacheable".
This
flag is only allowed if (1) is valid.
This results in the following behaviours:
- If the VMM creates the memslot with the cacheable attribute without
(1) being advertised, we fail.
- If the VMM creates the memslot without the cacheable attribute, we
map as NC, as it is today.
Is that OK though?
Now we have the MM page tables mapping this memory as cachable but KVM
and the guest is accessing it as non-cached.
I don't think we should allow this.
I thought ARM tried hard to avoid creating such mismatches? This is
why the pgprot flags were used to drive this, not an opt-in flag. To
prevent userspace from forcing a mismatch.
We have the vma->vm_page_prot when the memslot is added, so we could
use
this instead of additional KVM flags.
I thought we try to avoid messing with the VMA when adding memslots;
because
KVM_CAP_SYNC_MMU allows user space for changing the VMAs afterwards
without
changing the memslot?
Any checks on the VMA at memslot creation is done out of courtesy to
userspace so it 'fails fast'. We repeat checks on the VMA at the time of
fault to handle userspace twiddling VMAs behind our back.
yes, I think it is better to add cachable attribute in memslot, it can
be checked on the VMA at memslot creation. Also cache attribute can be
abstracted with cachable/uc/wc type rather than detailed arch specified.
Sorry, I do not state this clearly. My meaning is tor add cachable
attribute in memslot. It is acquired from prot of VMA at memslot
creation, and checked at S2 page fault fastpath.
So it is unnecessary to find vma at S2 page fault fastpath, only memslot
is enough. And cache attribute write-combined should be added also since
some GPU memory can be mapped with WC attribute in vfio-gpu driver.
Regards
Bibo Mao
Regards
Bibo Mao
VM_MTE_ALLOWED is an example of this.
Thanks,
Oliver
