On 11/25/2015 08:14 PM, Paolo Bonzini wrote:
On 25/11/2015 17:55, Andrey Smetanin wrote:
+ gpa = synic->msg_page & PAGE_MASK;
+ page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
+ if (is_error_page(page)) {
+ vcpu_err(vcpu, "Hyper-V SynIC can't get msg page, gpa 0x%llx\n",
+ gpa);
+ return;
+ }
+ msg_page = kmap_atomic(page);
But the message page is not being pinned, is it?
Actually I don't know anything about pinning.
Is it pinning against page swapping ?
Yes. Unless the page is pinned, kmap_atomic can fail.
kmap_atomic() can't fail for a valid page struct. Does
kvm_vcpu_gfn_to_page() can provide invalid page(swapped page) struct
which may pass is_error_page(page) check but can leads to incorrect
behavior inside kmap_atomic()?
However, I don't think that kvm_hv_notify_acked_sint is called from
atomic context. It is only called from apic_set_eoi. Could you just
use kvm_vcpu_write_guest_page?
In this case I can use kvm_vcpu_write_guest_page(), but in the 'PATCH v1
7/7' I do the same page mapping method to sync_cmpxchg() at guest
message page address to exclusively acquire message page slot(see
synic_deliver_msg()). So we need some method to map and access
atomically memory of guest page in KVM. Does any method to pin and map
guest page in kernel exists? Or should we use mlock() for this page in
QEMU part ?
By the way, do you need to do this also in kvm_get_apic_interrupt, for
auto EOI interrupts?
No we don't need this because in case of auto EOI interrupts, if
->msg_pending was set, host will receive HV_X64_MSR_EOM write request
which calls kvm_hv_notify_acked_sint().
Thanks,
Paolo
Could you please clarify and provide an API to use in this case ?
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
