[PATCH 2/2] powerpc/mce: Do not poison the memory using guest effective addr

Ganesh Goudar ganeshgr at linux.ibm.com
Mon Apr 27 16:13:08 AEST 2020


As of now, if we hit UE due to memory failure in guest, host MCE
handler tries to find the pfn or physical address where the memory
error occurred using guest effective address and uses that pfn to
poison the memory, which is not right.

If we hit UE in guest, do not try to find pfn in host and thereby
avoid poisoning the memory in host.

Reviewed-by: Mahesh Salgaonkar <mahesh at linux.vnet.ibm.com>
Signed-off-by: Ganesh Goudar <ganeshgr at linux.ibm.com>
---
 arch/powerpc/kernel/mce_power.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/arch/powerpc/kernel/mce_power.c b/arch/powerpc/kernel/mce_power.c
index 143e79450e93..4e541dd3af0d 100644
--- a/arch/powerpc/kernel/mce_power.c
+++ b/arch/powerpc/kernel/mce_power.c
@@ -471,6 +471,10 @@ static int mce_handle_ierror(struct pt_regs *regs,
 				table[i].error_type == MCE_ERROR_TYPE_UE) {
 				unsigned long pfn;
 
+#ifdef CONFIG_KVM_BOOK3S_HANDLER
+				if (get_paca()->kvm_hstate.in_guest)
+					return handled;
+#endif
 				if (get_paca()->in_mce < MAX_MCE_DEPTH) {
 					pfn = addr_to_pfn(regs, regs->nip);
 					if (pfn != ULONG_MAX) {
@@ -515,6 +519,7 @@ static int mce_handle_derror(struct pt_regs *regs,
 		 */
 		if (found)
 			continue;
+		found = 1;
 
 		/* now fill in mce_error_info */
 		mce_err->error_type = table[i].error_type;
@@ -528,6 +533,10 @@ static int mce_handle_derror(struct pt_regs *regs,
 			*addr = regs->dar;
 		else if (mce_err->sync_error &&
 				table[i].error_type == MCE_ERROR_TYPE_UE) {
+#ifdef CONFIG_KVM_BOOK3S_HANDLER
+			if (get_paca()->kvm_hstate.in_guest)
+				continue;
+#endif
 			/*
 			 * We do a maximum of 4 nested MCE calls, see
 			 * kernel/exception-64s.h
@@ -536,7 +545,6 @@ static int mce_handle_derror(struct pt_regs *regs,
 				mce_find_instr_ea_and_phys(regs, addr,
 							   phys_addr);
 		}
-		found = 1;
 	}
 
 	if (found)
-- 
2.17.2



More information about the Linuxppc-dev mailing list