mirror of
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
synced 2026-08-30 04:09:55 -04:00
2417218f2f
KVM caches MDCR_EL2 on a per-CPU basis in order to preserve the
configuration of MDCR_EL2.HPMN while running a guest. This is a bit
gross, since we're relying on some baked configuration rather than the
hardware definition of implemented counters.
Discover the number of implemented counters by reading PMCR_EL0.N
instead. This works because:
- In VHE the kernel runs at EL2, and N always returns the number of
counters implemented in hardware
- In {n,h}VHE, the EL2 setup code programs MDCR_EL2.HPMN with the EL2
view of PMCR_EL0.N for the host
Lastly, avoid traps under nested virtualization by saving PMCR_EL0.N in
host data.
Tested-by: James Clark <james.clark@linaro.org>
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Link: https://lore.kernel.org/r/20241219224116.3941496-3-oliver.upton@linux.dev
Signed-off-by: Marc Zyngier <maz@kernel.org>
22 lines
399 B
C
22 lines
399 B
C
// SPDX-License-Identifier: GPL-2.0-only
|
|
/*
|
|
* Copyright (C) 2015 - ARM Ltd
|
|
* Author: Marc Zyngier <marc.zyngier@arm.com>
|
|
*/
|
|
|
|
#include <hyp/debug-sr.h>
|
|
|
|
#include <linux/kvm_host.h>
|
|
|
|
#include <asm/kvm_hyp.h>
|
|
|
|
void __debug_switch_to_guest(struct kvm_vcpu *vcpu)
|
|
{
|
|
__debug_switch_to_guest_common(vcpu);
|
|
}
|
|
|
|
void __debug_switch_to_host(struct kvm_vcpu *vcpu)
|
|
{
|
|
__debug_switch_to_host_common(vcpu);
|
|
}
|