On Wed, Apr 08, 2020 at 03:01:58PM +0200, Thomas Gleixner wrote:
And it comes with restrictions:
The Do Other Stuff event can only be delivered when guest IF=1. If guest IF=0 then the host has to suspend the guest until the situation is resolved. The 'Situation resolved' event must also wait for a guest IF=1 slot.
Moo, can we pretty please already kill that ALWAYS and IF nonsense? That results in that terrifyingly crap HLT loop. That needs to die with extreme prejudice.
So the host only inject these OMFG_DOS things when the guest is in luserspace -- which it can see in the VMCS state IIRC. And then using #VE for the make-it-go signal is far preferred over the currentl #PF abuse.
Page-not-present async page faults are almost a perfect match for the hardware use of #VE (and it might even be possible to let the processor deliver the exceptions). There are other advantages:
- the only real problem with using #PF (with or without
KVM_ASYNC_PF_SEND_ALWAYS) seems to be the NMI reentrancy issue, which would not be there for #VE.
- #VE are combined the right way with other exceptions (the
benign/contributory/pagefault stuff)
- adjusting KVM and Linux to use #VE instead of #PF would be less than
100 lines of code.
If you just want to solve Viveks problem, then its good enough. I.e. the file truncation turns the EPT entries into #VE convertible entries and the guest #VE handler can figure it out. This one can be injected directly by the hardware, i.e. you don't need a VMEXIT.
That sounds like something that doesn't actually need the whole 'async'/do-something-else-for-a-while crap, right? It's a #PF trap from kernel space where we need to report fail.
If you want the opportunistic do other stuff mechanism, then #VE has exactly the same problems as the existing async "PF". It's not magicaly making that go away.
We need to somehow have the guest teach the host how to tell if it can inject that OMFG_DOS thing or not. Injecting it only to then instantly exit again is stupid and expensive.
Clearly we don't want to expose preempt_count and make that ABI, but is there some way we can push a snippet of code to the host that instructs the host how to determine if it can sleep or not? I realize that pushing actual x86 .text is a giant security problem, so perhaps a snipped of BPF that the host can verify, which it can run on the guest image ?
Make it a hard error (guest cpu dies) to inject the OMFG_DOS signal on a context that cannot put the task to sleep.