Re: [PATCH] mem_event: Allow emulating an instruction that caused a page fault

[Razvan Cojocaru <rzvncj@gmail.com>]

> Hi,

Hello Tim, thank you for taking the time to review this.

> I think there ought to be some other control to this: what if a single
> instruction accesses multiple pages, each of which would cause an access
> fault?  You only get a notification of the first one, so short of
> emulating the instruction yourself in userspace I don't know how you can
> decide that it's safe.

You're right, but I can't see how this case could be handled at all 
without lifting the restrictions, one page at a time. And that's 
precisely what this patch aims to make unnecessary. I can't see a way 
around it (not while the emulation support is limited to 
hvm_emulate_one() and hvm_emulate_one_nowrite()).

> This function always operates on the currently scheduled vcpu, so you
> don't need to pass a cpu-user-regs struct all the way down the stack --
> you can just use guest_cpu_user_regs() here.

Of course. Thank you.

> I don't think this is necessary: you only need to add a field to thuiis
> file if you'll be using it from assembly code.

I thought I'd be polite and add it anyway, in case somebody will want to 
use it later. Was that pollution? I'll remove it.

>> +#define MEM_EVENT_FLAG_EMULATE      (1 << 5)
>
> Please add a comment saying what this flag does.  I know the rest of
> this code is poorly commented, but let's try to make thing better as we
> go. :)

I will.

Thank you,
Razvan Cojocaru