Currently, rebooting a pseries nested qemu-kvm guest (L2) results in
below error as L1 qemu sends PVR value 'arch_compat' == 0 via
ppc_set_compat ioctl. This triggers a condition failure in
kvmppc_set_arch_compat() resulting in an EINVAL.
qemu-system-ppc64: Unable to set CPU compatibility mode in KVM: Invalid
argument
Also, a value of 0 for arch_compat generally refers the default
compatibility of the host. But, arch_compat, being a Guest Wide Element
in nested API v2, cannot be set to 0 in GSB as PowerVM (L0) expects a
non-zero value. A value of 0 triggers a kernel trap during a reboot and
consequently causes it to fail:
[ 22.106360] reboot: Restarting system
KVM: unknown exit, hardware reason ffea
NIP 0100 LR fe44 CTR XER
20040092 CPU#0
MSR 1000 HID0 HF 6c00 iidx 3 didx 3
TB DECR 0
GPR00 c2a8c300 7fe0
GPR04 1002 82803033
GPR08 0a00 0004 2fff
GPR12 c2e1 000105639200 0004
GPR16 00010563a090
GPR20 000105639e20 0001056399c8 7fffe54abab0 000105639288
GPR24 0001 0001
GPR28 c2b30840
CR [ - - - - - - - - ] RES 000@
SRR0 SRR1 PVR 00800200 VRSAVE
SPRG0 SPRG1 SPRG2 SPRG3
SPRG4 SPRG5 SPRG6 SPRG7
HSRR0 HSRR1
CFAR
LPCR 00020400
PTCR DAR DSISR
kernel:trap=0xffea | pc=0x100 | msr=0x1000
This patch updates kvmppc_set_arch_compat() to use the host PVR value if
'compat_pvr' == 0 indicating that qemu doesn't want to enforce any
specific PVR compat mode.
The relevant part of the code might need a rework if PowerVM implements
a support for `arch_compat == 0` in nestedv2 API.
Fixes: 19d31c5f1157 ("KVM: PPC: Add support for nestedv2 guests")
Reviewed-by: Aneesh Kumar K.V (IBM)
Reviewed-by: Vaibhav Jain
Signed-off-by: Amit Machhiwal
---
Changes v3 -> v4:
- Moved part of a code comment around implementation of `arch_compat
== 0` in PowerVM to the patch description based on an off mailing
list discussion
Changes v2 -> v3:
- Vaibhav: Use a 'break' instead of switch-case fallthrough
- v2:
https://lore.kernel.org/all/20240205132607.2776637-1-amach...@linux.ibm.com/
Changes v1 -> v2:
- Added descriptive error log in the patch description when
`arch_compat == 0` passed in GSB
- Added a helper function for PCR to capabilities mapping
- Added relevant comments around the changes being made
- v1:
https://lore.kernel.org/lkml/20240118095653.2588129-1-amach...@linux.ibm.com/
arch/powerpc/kvm/book3s_hv.c | 26 --
arch/powerpc/kvm/book3s_hv_nestedv2.c | 20 ++--
2 files changed, 42 insertions(+), 4 deletions(-)
diff --git a/arch/powerpc/kvm/book3s_hv.c b/arch/powerpc/kvm/book3s_hv.c
index 52427fc2a33f..0b921704da45 100644
--- a/arch/powerpc/kvm/book3s_hv.c
+++ b/arch/powerpc/kvm/book3s_hv.c
@@ -391,6 +391,24 @@ static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32
pvr)
/* Dummy value used in computing PCR value below */
#define PCR_ARCH_31(PCR_ARCH_300 << 1)
+static inline unsigned long map_pcr_to_cap(unsigned long pcr)
+{
+ unsigned long cap = 0;
+
+ switch (pcr) {
+ case PCR_ARCH_300:
+ cap = H_GUEST_CAP_POWER9;
+ break;
+ case PCR_ARCH_31:
+ cap = H_GUEST_CAP_POWER10;
+ break;
+ default:
+ break;
+ }
+
+ return cap;
+}
+
static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
{
unsigned long host_pcr_bit = 0, guest_pcr_bit = 0, cap = 0;
@@ -424,11 +442,9 @@ static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu,
u32 arch_compat)
break;
case PVR_ARCH_300:
guest_pcr_bit = PCR_ARCH_300;
- cap = H_GUEST_CAP_POWER9;
break;
case PVR_ARCH_31:
guest_pcr_bit = PCR_ARCH_31;
- cap = H_GUEST_CAP_POWER10;
break;
default:
return -EINVAL;
@@ -440,6 +456,12 @@ static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu,
u32 arch_compat)
return -EINVAL;
if (kvmhv_on_pseries() &&