On Mon, Nov 11, 2019 at 04:28:06PM +1100, Paul Mackerras wrote:
On Mon, Nov 04, 2019 at 09:47:58AM +0530, Bharata B Rao wrote:
quoted
Add support for reset of secure guest via a new ioctl KVM_PPC_SVM_OFF.
This ioctl will be issued by QEMU during reset and includes the
the following steps:
- Ask UV to terminate the guest via UV_SVM_TERMINATE ucall
- Unpin the VPA pages so that they can be migrated back to secure
side when guest becomes secure again. This is required because
pinned pages can't be migrated.
Unpinning the VPA pages is normally handled during VM reset by QEMU
doing set_one_reg operations to set the values for the
KVM_REG_PPC_VPA_ADDR, KVM_REG_PPC_VPA_SLB and KVM_REG_PPC_VPA_DTL
pseudo-registers to zero. Is there some reason why this isn't
happening for a secure VM, and if so, what is that reason?
If it is happening, then why do we need to unpin the pages explicitly
here?
We were observing these VPA pages still remaining pinned during
reset and hence subsequent paging-in of these pages were failing.
Unpinning them fixed the problem.
I will investigate and get back on why exactly these pages weren't
gettting unpinned normally as part of reset.
quoted
- Reinitialize guest's partitioned scoped page tables. These are
freed when guest becomes secure (H_SVM_INIT_DONE)
It doesn't seem particularly useful to me to free the partition-scoped
page tables when the guest becomes secure, and it feels like it makes
things more fragile. If you don't free them then, then you don't need
to reallocate them now.
Sure, I will not free them in the next version.
Regards,
Bharata.