Virtualization - Jun 2021 - [PATCH v3 4/7] x86/sev-es: Run #VC handler in plain IRQ state

From: Joerg Roedel <jroedel at suse.de>

Use irqentry_enter() and irqentry_exit() to track the runtime state of
the #VC handler. The reason it ran in NMI mode was solely to make sure
nothing interrupts the handler while the GHCB is in use.

This is handled now in sev_es_get/put_ghcb() directly, so there is no
reason the #VC handler can not run in normal IRQ mode and enjoy the
benefits like being able to send signals.

Fixes: 62441a1fb532 ("x86/sev-es: Correctly track IRQ states in runtime #VC
handler")
Signed-off-by: Joerg Roedel <jroedel at suse.de>
---
 arch/x86/kernel/sev.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/arch/x86/kernel/sev.c b/arch/x86/kernel/sev.c
index 2a922d1b03c8..b563fb747aed 100644
--- a/arch/x86/kernel/sev.c
+++ b/arch/x86/kernel/sev.c
@@ -1354,8 +1354,7 @@ DEFINE_IDTENTRY_VC_SAFE_STACK(exc_vmm_communication)
 		return;
 	}
 
-	irq_state = irqentry_nmi_enter(regs);
-	lockdep_assert_irqs_disabled();
+	irq_state = irqentry_enter(regs);
 	instrumentation_begin();
 
 	/*
@@ -1408,7 +1407,7 @@ DEFINE_IDTENTRY_VC_SAFE_STACK(exc_vmm_communication)
 
 out:
 	instrumentation_end();
-	irqentry_nmi_exit(regs, irq_state);
+	irqentry_exit(regs, irq_state);
 
 	return;
 
-- 
2.31.1

On Tue, Jun 08, 2021 at 11:54:36AM +0200, Joerg Roedel
wrote:
> From: Joerg Roedel <jroedel at suse.de>
> 
> Use irqentry_enter() and irqentry_exit() to track the runtime state of
> the #VC handler. The reason it ran in NMI mode was solely to make sure
> nothing interrupts the handler while the GHCB is in use.
> 
> This is handled now in sev_es_get/put_ghcb() directly, so there is no
> reason the #VC handler can not run in normal IRQ mode and enjoy the
> benefits like being able to send signals.
You sure?

So #VC cannot happen with IRQs disabled?

	raw_spin_lock_irq(&my_lock);
	<#VC>
		raw_spin_lock_irqsave(&my_lock); // whoopsie

Every exception that can happen with IRQs disabled must be NMI like.

Again, what you seem to want is to split the handler in a from-user and
from-kernel way, just like we did with #DB and MCE. See how
exc_debug_user() is IRQ-like and can send signals, while
exc_debug_kernel() is NMI like and can not.