On Thu, Jul 29, 2021 at 11:19:08AM +0200, Paolo Bonzini wrote:

On 29/07/21 05:00, Yu Zhang wrote:

quoted

quoted

I have a few questions about these unnecessary tear-downs during boot:
1. How many teardowns did you observe, and how many different roles
did they represent? Just thrashing between two roles, or 12 different
roles?

I saw 106 reloadings of the root TDP. Among them, 14 are caused by memslot
changes. Remaining ones are caused by the context reset from CR0/CR4/EFER
changes(85 for CR0 changes).

Possibly because CR0/CR4/EFER are changed multiple times on SMM entry (to go
from real mode to protected mode to 32-bit to 64-bit)?  But most of those
page tables should be very very small; they probably have only one page per
level.  The SMM page tables are very small too, the only one that is really
expensive to rebuild is the main non-SMM EPT.

Thanks Paolo. 

Well, I did not see any SMM entry in the whole test. And most resetings
are due to CR0 changes in OVMF(74 out of 85), the rest are from guest
kernel initialization stage.

As expected, the number of SPs used are fairly small - about 5 - 10 for
each TDP tree. For legcy TDP, since only 4 different TDP trees are built(
the ones caused by memslot zapping are not counted). The total number of
SPs are only 30+.

B.R.
Yu

Paolo