Linux Virtualization - Jun 2020 - [PATCH v3 47/75] x86/sev-es: Add Runtime #VC Exception Handler

On Tue, Apr 28, 2020 at 05:16:57PM +0200, Joerg Roedel
wrote:
> diff --git a/arch/x86/kernel/sev-es.c b/arch/x86/kernel/sev-es.c
> index a4fa7f351bf2..bc3a58427028 100644
> --- a/arch/x86/kernel/sev-es.c
> +++ b/arch/x86/kernel/sev-es.c
> @@ -10,6 +10,7 @@
>  #include <linux/sched/debug.h>	/* For show_regs() */
>  #include <linux/percpu-defs.h>
>  #include <linux/mem_encrypt.h>
> +#include <linux/lockdep.h>
>  #include <linux/printk.h>
>  #include <linux/mm_types.h>
>  #include <linux/set_memory.h>
> @@ -25,7 +26,7 @@
>  #include <asm/insn-eval.h>
>  #include <asm/fpu/internal.h>
>  #include <asm/processor.h>
> -#include <asm/trap_defs.h>
> +#include <asm/traps.h>
>  #include <asm/svm.h>
>  
>  /* For early boot hypervisor communication in SEV-ES enabled guests */
> @@ -46,10 +47,26 @@ struct sev_es_runtime_data {
>  
>  	/* Physical storage for the per-cpu IST stacks of the #VC handler */
>  	struct vmm_exception_stacks vc_stacks __aligned(PAGE_SIZE);
> +
> +	/* Reserve on page per CPU as backup storage for the unencrypted GHCB */
		  one
> +	struct ghcb backup_ghcb;
I could use some text explaining what those backups are for?
> +	/*
> +	 * Mark the per-cpu GHCBs as in-use to detect nested #VC exceptions.
> +	 * There is no need for it to be atomic, because nothing is written to
> +	 * the GHCB between the read and the write of ghcb_active. So it is safe
> +	 * to use it when a nested #VC exception happens before the write.
> +	 */
Looks liks that is that text... support for nested #VC exceptions.
I'm sure this has come up already but why do we even want to support
nested #VCs? IOW, can we do without them first or are they absolutely
necessary?

I'm guessing VC exceptions inside the VC handler but what are the
sensible use cases?

Thx.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette

On Sat, May 23, 2020 at 09:59:24AM +0200, Borislav Petkov
wrote:
> On Tue, Apr 28, 2020 at 05:16:57PM +0200, Joerg Roedel wrote:
> > +	/*
> > +	 * Mark the per-cpu GHCBs as in-use to detect nested #VC exceptions.
> > +	 * There is no need for it to be atomic, because nothing is written
to
> > +	 * the GHCB between the read and the write of ghcb_active. So it is
safe
> > +	 * to use it when a nested #VC exception happens before the write.
> > +	 */
> 
> Looks liks that is that text... support for nested #VC exceptions.
> I'm sure this has come up already but why do we even want to support
> nested #VCs? IOW, can we do without them first or are they absolutely
> necessary?
> 
> I'm guessing VC exceptions inside the VC handler but what are the
> sensible use cases?
The most important use-case is #VC->NMI->#VC. When an NMI hits while the
#VC handler uses the GHCB and the NMI handler causes another #VC, then
the contents of the GHCB needs to be backed up, so that it doesn't
destroy the GHCB contents of the first #VC handling path.


Regards,

	Joerg

On Sat, May 23, 2020 at 09:59:24AM +0200, Borislav Petkov
wrote:
> On Tue, Apr 28, 2020 at 05:16:57PM +0200, Joerg Roedel wrote:
> > +	struct ghcb backup_ghcb;
> 
> I could use some text explaining what those backups are for?
I added another comment above that line to better explain why it is
needed.


	Joerg

On Thu, Jun 11, 2020 at 01:48:31PM +0200, Joerg Roedel
wrote:
> On Sat, May 23, 2020 at 09:59:24AM +0200, Borislav Petkov wrote:
> > On Tue, Apr 28, 2020 at 05:16:57PM +0200, Joerg Roedel wrote:
> > > +	/*
> > > +	 * Mark the per-cpu GHCBs as in-use to detect nested #VC
exceptions.
> > > +	 * There is no need for it to be atomic, because nothing is
written to
> > > +	 * the GHCB between the read and the write of ghcb_active. So
it is safe
> > > +	 * to use it when a nested #VC exception happens before the
write.
> > > +	 */
> > 
> > Looks liks that is that text... support for nested #VC exceptions.
> > I'm sure this has come up already but why do we even want to
support
> > nested #VCs? IOW, can we do without them first or are they absolutely
> > necessary?
> > 
> > I'm guessing VC exceptions inside the VC handler but what are the
> > sensible use cases?
> 
> The most important use-case is #VC->NMI->#VC. When an NMI hits while
the
> #VC handler uses the GHCB and the NMI handler causes another #VC, then
> the contents of the GHCB needs to be backed up, so that it doesn't
> destroy the GHCB contents of the first #VC handling path.
Isn't it possible for the #VC handler to hit a #PF, e.g. on copy_from_user()
in insn_fetch_from_user()?  If that happens, what prevents the #PF handler
from hitting a #VC?  AIUI, do_vmm_communication() panics if the backup GHCB
is already in use, e.g. #VC->#PF->#VC->NMI->#VC would be fatal.

On Thu, Jun 11, 2020 at 01:48:31PM +0200, Joerg Roedel
wrote:
> The most important use-case is #VC->NMI->#VC. When an NMI hits while
the
> #VC handler uses the GHCB and the NMI handler causes another #VC, then
> the contents of the GHCB needs to be backed up, so that it doesn't
> destroy the GHCB contents of the first #VC handling path.
That's a good example, please add it to the next version of the patch,
preferrably in a comment somewhere.

Thx.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette