Hi James,
On 10/24/20 3:54 AM, James Morse wrote:
On 18/08/2020 02:13, Gavin Shan wrote:
This renames user_mem_abort() to kvm_handle_user_mem_abort(), and
then exports it. The function will be used in asynchronous page fault
to populate a page table entry once the corresponding page is populated
from the backup device (e.g. swap partition):
* Parameter @fault_status is replace by @esr.
* Parameter @prefault is added
As the @esr is passed as parameter, not fetched from vCPU struct. This
also introduces the necessasry helpers in esr.h, to manupulate the @esr.
(Nit: necessary, manipulate)
Thanks for your comments. It will be fixed in next revision :)
The helpers defined in kvm_emulate.h reuses the newly added helper. This
shouldn't cause functional changes.
user_mem_abort() is deep in the the guts of KVM's arch code. I don't think this should be
exported. It must be called on the vcpu thread. It must be called under the VMs srcu
lock. There are also tricky interactions with whether the vcpu is loaded on this cpu or not...
I think it would be much simpler to always let the guest take the stage2-fault a second
time. This keeps the property that pages are only populate in response to a stage2 fault.
If the guest behaves, it will only schedule a task that accesses the page once its available.
The function is called with SRCU hold in this patchset. However, it seems
more reasonable to be called in vCPU thread. On the other hand, I'm worrying
about the performance degradation. I will evaluate it in next revision and
leave the function is called in vCPU thread if performance is acceptable.
Otherwise, I would like to leave it as of being :)
Thanks,
Gavin
_______________________________________________
kvmarm mailing list
kvmarm@xxxxxxxxxxxxxxxxxxxxx
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
