Re: [PATCH] vm_event: Record FS_BASE/GS_BASE during events

[Razvan Cojocaru <rcojocaru@bitdefender.com>]

On 02/11/2016 10:38 PM, Tamas K Lengyel wrote:
> 
> 
> On Thu, Feb 11, 2016 at 1:13 PM, Razvan Cojocaru
> <rcojocaru@bitdefender.com <mailto:rcojocaru@bitdefender.com>> wrote:
> 
>     On 02/11/2016 10:04 PM, Andrew Cooper wrote:
>     > On 11/02/16 20:00, Razvan Cojocaru wrote:
>     >> On 02/11/2016 09:55 PM, Andrew Cooper wrote:
>     >>> On 11/02/16 19:54, Razvan Cojocaru wrote:
>     >>>> On 02/11/2016 09:51 PM, Tamas K Lengyel wrote:
>     >>>>> While the public vm_event header specifies fs_base/gs_base as
>     registers that
>     >>>>> should be recorded for each event, that hasn't actually been
>     the case. In
>     >>>>> this patch we remedy the issue.
>     >>>>>
>     >>>>> Signed-off-by: Tamas K Lengyel <tlengyel@novetta.com
>     <mailto:tlengyel@novetta.com>>
>     >>>>> Cc: Razvan Cojocaru <rcojocaru@bitdefender.com
>     <mailto:rcojocaru@bitdefender.com>>
>     >>>>> Cc: Keir Fraser <keir@xen.org <mailto:keir@xen.org>>
>     >>>>> Cc: Jan Beulich <jbeulich@suse.com <mailto:jbeulich@suse.com>>
>     >>>>> Cc: Andrew Cooper <andrew.cooper3@citrix.com
>     <mailto:andrew.cooper3@citrix.com>>
>     >>>>> ---
>     >>>>>  xen/arch/x86/hvm/event.c | 9 ++++++++-
>     >>>>>  1 file changed, 8 insertions(+), 1 deletion(-)
>     >>>> Fair enough.
>     >>>>
>     >>>> Acked-by: Razvan Cojocaru <rcojocaru@bitdefender.com
>     <mailto:rcojocaru@bitdefender.com>>
>     >>> Oops.
>     >>>
>     >>> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com
>     <mailto:andrew.cooper3@citrix.com>>
>     >> This has actually been intentional, in that we've only needed those
>     >> fields for EPT events, and thought that not filling what's not needed
>     >> until it's needed would save a tiny bit of hypervisor processing
>     time.
>     >> They are being filled in only for page fault events at the moment.
>     >>
>     >> I believe it's been discussed at the time. We still don't need those
>     >> coming with the events that use hvm_event_fill_regs(), but if Tamas
>     >> needs them then by all means.
>     >
>     > The public header file does suggest that all of vm_event_regs_x86 will
>     > be complete.  Are there any other fields currently missing?
> 
>     There are. p2m_vm_event_fill_regs() fills everything in (in
>     xen/arch/x86/mm/p2m.c). hvm_event_fill_regs() still does not, even after
>     Tamas' patch.
> 
> 
> Ah, that makes sense. Yea, I would prefer if all registers would get
> filled in for all events so I'll just consolidate these two functions
> into one.

Right, but please be careful and test that you get correct values with
all events (page fault events + the others), I remember that for some
reason I needed to use different ways to get at the same values in
p2m_vm_event_fill_regs() and hvm_event_fill_regs().

For example, p2m_vm_event_fill_regs() does:

hvm_funcs.save_cpu_ctxt(curr, &ctxt);
req->data.regs.x86.cr0 = ctxt.cr0;

and hvm_event_fill_regs() does:

req->data.regs.x86.cr0 = curr->arch.hvm_vcpu.guest_cr[0];

I don't remember exactly why I had to do that at the time, but I do
recall it being necessary.


Thanks,
Razvan