On 2022/10/11 21:57, Dave Hansen wrote:
> That honestly doesn't help me understand what's going on here.
>
> The changelog and subject are spending far too much time explaining the
> low level details of what is being done (like "add a union") and far too
> little time explaining what it _means_.
Thanks for your comments, Dave.
I added more descriptions for motivation and reduced explanations
for details in changelog as below, could you help me review if it is
easier to understand now?
When a page triggers a machine check, it only reports the PFN. But
in order to inject #MC into VM through hypervisor, the virtual
address of virtual EPC page is required. To avoid introduce a new
dedicated structure to track it, repurpose the 'encl_owner' field as
it is useless in virtualization case, so that arch_memory_failure()
can retrieve the virtual address easily and attach it to the extra
info of SIGBUS which will be sent to hypervisor.
The union can prevent playing the casting games while using it. And
add a new EPC page flag - SGX_EPC_PAGE_KVM_GUEST to interpret the
meaning of the field.
The changes take place while allocating an EPC page to a KVM guests.
After applying it, the reversed relationship from an EPC page to
virtual EPC doesn't exist, in practice, such relationship is
useless. Beyond that, other behaviors have not been changed.
Furthermore, may I change the subject as below to highlight the
motivation?
x86/sgx: track virtual address of virtual EPC page for injecting #MC
into VM
Best Regards,
Zhiquan
