[PATCH] x86/CPU/AMD: Don't print reset reason in VMs

[PATCH] x86/CPU/AMD: Don't print reset reason in VMs

[Jason Andryuk]

print_s5_reset_status_mmio() expects to read FCH MMIO on baremetal.  VMs
don't have emulated FCH registers, so there is nothing to check.  Add an
early exit when running on a hypervisor.

The real motivation is Xen PV domUs can't boot on Zen CPUs.  PV domUs
PTE contain system physical addresses - they don't have concept of guest
physical addresses.  domUs can only map physical addresses they are
assigned, so a domU fails to map the FCH MMIO addresses.  However
set_pte() does not return failure, so ioremap() returns a pointer while
the PTEs are unpopulated.  print_s5_reset_status_mmio() faults when
reading through the pointer, breaking the boot.  This change fixes Xen
PV domUs booting on Zen CPUs.

dom0 can map the FCH MMIO, but it has X86_FEATURE_HYPERVISOR set.  With
this change, it no longer prints the reset reason.

Fixes: ab8131028710 ("x86/CPU/AMD: Print the reason for the last reset")
Cc: stable@vger.kernel.org
Suggested-by: Borislav Petkov <bp@alien8.de>
Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
---
dom0 could work, but that would require a xen_initial_domain() check.

I've wondered about a different approach:  We can't directly check for a
device for the FCH_PM_BASE memory, but maybe we could check for a PCI
ISA Bridge with PCI_VENDOR_ID_AMD.  That might be close enough to
indicate real hardware?  But it would still be a heuristic, so I'm not
sure how much better that would be.
---
 arch/x86/kernel/cpu/amd.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/x86/kernel/cpu/amd.c b/arch/x86/kernel/cpu/amd.c
index 40463bad1078..580ccc2ce5fc 100644
--- a/arch/x86/kernel/cpu/amd.c
+++ b/arch/x86/kernel/cpu/amd.c
@@ -1357,7 +1357,8 @@ static __init int print_s5_reset_status_mmio(void)
 	u32 value;
 	int i;
 
-	if (!cpu_feature_enabled(X86_FEATURE_ZEN))
+	if (!cpu_feature_enabled(X86_FEATURE_ZEN) ||
+	    cpu_feature_enabled(X86_FEATURE_HYPERVISOR))
 		return 0;
 
 	addr = ioremap(FCH_PM_BASE + FCH_PM_S5_RESET_STATUS, sizeof(value));
-- 
2.34.1

Re: [PATCH] x86/CPU/AMD: Don't print reset reason in VMs

[Juergen Gross]

[-- Attachment #1.1: Type: text/plain, Size: 1189 bytes --]

On 04.11.26 02:30, Jason Andryuk wrote:
> print_s5_reset_status_mmio() expects to read FCH MMIO on baremetal.  VMs
> don't have emulated FCH registers, so there is nothing to check.  Add an
> early exit when running on a hypervisor.
> 
> The real motivation is Xen PV domUs can't boot on Zen CPUs.  PV domUs
> PTE contain system physical addresses - they don't have concept of guest
> physical addresses.  domUs can only map physical addresses they are
> assigned, so a domU fails to map the FCH MMIO addresses.  However
> set_pte() does not return failure, so ioremap() returns a pointer while
> the PTEs are unpopulated.  print_s5_reset_status_mmio() faults when
> reading through the pointer, breaking the boot.  This change fixes Xen
> PV domUs booting on Zen CPUs.
> 
> dom0 can map the FCH MMIO, but it has X86_FEATURE_HYPERVISOR set.  With
> this change, it no longer prints the reset reason.
> 
> Fixes: ab8131028710 ("x86/CPU/AMD: Print the reason for the last reset")
> Cc: stable@vger.kernel.org
> Suggested-by: Borislav Petkov <bp@alien8.de>
> Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
Reviewed-by: Juergen Gross <jgross@suse.com>


Juergen

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 495 bytes --]