[PATCH] x86/CPU/AMD: fix booting on Xen PV domU

From: Marek Marczykowski-Górecki

Date: Sun Nov 15 2026 - 20:41:28 EST


Reading the FCH_PM_S5_RESET_STATUS register doesn't make sense in a VM,
as it's about reporting last hardware reset reason and VM has no access
to it. On most hypervisors, reading it will simply return all-ones, but
unfortunately Xen PV domU is (again) special and trying to read not
allowed MMIO space results in a panic:

1 of 1 multicall(s) failed: cpu 1
call 1: op=1 arg=[ffff88801851b010] result=-22
BUG: unable to handle page fault for address: ffffc9004002d3c0
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
PGD 18f1b067 P4D 18f1b067 PUD 3e8f067 PMD 3e90067 PTE 0
Oops: Oops: 0000 [#1] SMP NOPTI
CPU: 0 UID: 0 PID: 1 Comm: swapper/0 Tainted: G W 6.18.62-1.qubes.28.fc41.x86_64 #1 PREEMPT(full)
Tainted: [W]=WARN
RIP: e030:ioread32+0x32/0x60
Code: ff 03 00 77 25 48 81 ff 00 00 01 00 76 08 89 fa ed e9 72 04 8c 00 8b 05 cc 62 88 01 85 c0 75 11 b8 ff ff ff ff e9 5e 04 8c 00 <8b> 07 e9 57 04 8c 00 83 e8 01 48 89 fe 48 c7 c2 67 db 14 82 48 c7
RSP: e02b:ffffc9004000be00 EFLAGS: 00010292
RAX: 43fa5d3f578e48f3 RBX: 928a59081d78a1d7 RCX: 00000000fed803c3
RDX: 00000000039fffff RSI: 00000000fed803c0 RDI: ffffc9004002d3c0
RBP: ffffc9004000be38 R08: 0000000000000000 R09: 00000000000fed80
R10: 00000000000fed80 R11: 0720072007200720 R12: ffff888005118800
R13: ffffc9004002d3c0 R14: 000000000000017a R15: ffffffff832a78cc
FS: 0000000000000000(0000) GS:ffff888094f59000(0000) knlGS:0000000000000000
CS: e030 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: ffffc9004002d3c0 CR3: 000000000242c000 CR4: 0000000000050660
Call Trace:
<TASK>
print_s5_reset_status_mmio+0x75/0x1c0
? __pfx_print_s5_reset_status_mmio+0x10/0x10
do_one_initcall+0x5a/0x310
do_initcalls+0x1ef/0x240
kernel_init_freeable+0x18a/0x210
? __pfx_kernel_init+0x10/0x10
kernel_init+0x1a/0x140
ret_from_fork+0x158/0x170
? __pfx_kernel_init+0x10/0x10
ret_from_fork_asm+0x1a/0x30
</TASK>
Modules linked in:
CR2: ffffc9004002d3c0

Fix this by skipping the read when running in a VM, except Xen dom0
(which technically is a VM, but does have full access to the hardware).

Signed-off-by: Marek Marczykowski-Górecki <marmarek@xxxxxxxxxxxxxxxxxxxxxx>
---
arch/x86/kernel/cpu/amd.c | 5 +++++
1 file changed, 5 insertions(+)

diff --git a/arch/x86/kernel/cpu/amd.c b/arch/x86/kernel/cpu/amd.c
index 15286b71e9bf..fe3fe545608e 100644
--- a/arch/x86/kernel/cpu/amd.c
+++ b/arch/x86/kernel/cpu/amd.c
@@ -26,6 +26,7 @@
#include <asm/resctrl.h>
#include <asm/msr.h>
#include <asm/sev.h>
+#include <xen/xen.h>

#ifdef CONFIG_X86_64
# include <asm/mmconfig.h>
@@ -1370,6 +1371,10 @@ static __init int print_s5_reset_status_mmio(void)
if (!cpu_feature_enabled(X86_FEATURE_ZEN))
return 0;

+ if (cpu_feature_enabled(X86_FEATURE_HYPERVISOR) &&
+ !xen_initial_domain())
+ return 0;
+
addr = ioremap(FCH_PM_BASE + FCH_PM_S5_RESET_STATUS, sizeof(value));
if (!addr)
return 0;
--
2.56.0