thebeanogamer / rpms / qemu-kvm

Forked from rpms/qemu-kvm 5 months ago
Clone
1be5c7
From 688c9f386635544dbc468171a32fbc84f0c9224e Mon Sep 17 00:00:00 2001
1be5c7
From: Paolo Bonzini <pbonzini@redhat.com>
1be5c7
Date: Fri, 18 Mar 2022 16:23:47 +0100
1be5c7
Subject: [PATCH 12/24] target/i386: kvm: do not access uninitialized variable
1be5c7
 on older kernels
1be5c7
1be5c7
RH-Author: Paul Lai <plai@redhat.com>
1be5c7
RH-MergeRequest: 176: Enable KVM AMX support
1be5c7
RH-Commit: [12/13] 776fac1e7d1aa16ec5f4d99ddad3039eab8212af
1be5c7
RH-Bugzilla: 1916415
1be5c7
RH-Acked-by: Cornelia Huck <cohuck@redhat.com>
1be5c7
RH-Acked-by: Igor Mammedov <imammedo@redhat.com>
1be5c7
RH-Acked-by: Paolo Bonzini <pbonzini@redhat.com>
1be5c7
1be5c7
KVM support for AMX includes a new system attribute, KVM_X86_XCOMP_GUEST_SUPP.
1be5c7
Commit 19db68ca68 ("x86: Grant AMX permission for guest", 2022-03-15) however
1be5c7
did not fully consider the behavior on older kernels.  First, it warns
1be5c7
too aggressively.  Second, it invokes the KVM_GET_DEVICE_ATTR ioctl
1be5c7
unconditionally and then uses the "bitmask" variable, which remains
1be5c7
uninitialized if the ioctl fails.  Third, kvm_ioctl returns -errno rather
1be5c7
than -1 on errors.
1be5c7
1be5c7
While at it, explain why the ioctl is needed and KVM_GET_SUPPORTED_CPUID
1be5c7
is not enough.
1be5c7
1be5c7
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
1be5c7
(cherry picked from commit 3ec5ad40081b14af28496198b4d08dbe13386790)
1be5c7
Signed-off-by: Paul Lai <plai@redhat.com>
1be5c7
---
1be5c7
 target/i386/kvm/kvm.c | 17 +++++++++++++----
1be5c7
 1 file changed, 13 insertions(+), 4 deletions(-)
1be5c7
1be5c7
diff --git a/target/i386/kvm/kvm.c b/target/i386/kvm/kvm.c
1be5c7
index b1128b0e07..bd439e56ad 100644
1be5c7
--- a/target/i386/kvm/kvm.c
1be5c7
+++ b/target/i386/kvm/kvm.c
1be5c7
@@ -409,6 +409,12 @@ uint32_t kvm_arch_get_supported_cpuid(KVMState *s, uint32_t function,
1be5c7
         }
1be5c7
     } else if (function == 0xd && index == 0 &&
1be5c7
                (reg == R_EAX || reg == R_EDX)) {
1be5c7
+        /*
1be5c7
+         * The value returned by KVM_GET_SUPPORTED_CPUID does not include
1be5c7
+         * features that still have to be enabled with the arch_prctl
1be5c7
+         * system call.  QEMU needs the full value, which is retrieved
1be5c7
+         * with KVM_GET_DEVICE_ATTR.
1be5c7
+         */
1be5c7
         struct kvm_device_attr attr = {
1be5c7
             .group = 0,
1be5c7
             .attr = KVM_X86_XCOMP_GUEST_SUPP,
1be5c7
@@ -417,13 +423,16 @@ uint32_t kvm_arch_get_supported_cpuid(KVMState *s, uint32_t function,
1be5c7
 
1be5c7
         bool sys_attr = kvm_check_extension(s, KVM_CAP_SYS_ATTRIBUTES);
1be5c7
         if (!sys_attr) {
1be5c7
-            warn_report("cannot get sys attribute capabilities %d", sys_attr);
1be5c7
+            return ret;
1be5c7
         }
1be5c7
 
1be5c7
         int rc = kvm_ioctl(s, KVM_GET_DEVICE_ATTR, &attr);
1be5c7
-        if (rc == -1 && (errno == ENXIO || errno == EINVAL)) {
1be5c7
-            warn_report("KVM_GET_DEVICE_ATTR(0, KVM_X86_XCOMP_GUEST_SUPP) "
1be5c7
-                        "error: %d", rc);
1be5c7
+        if (rc < 0) {
1be5c7
+            if (rc != -ENXIO) {
1be5c7
+                warn_report("KVM_GET_DEVICE_ATTR(0, KVM_X86_XCOMP_GUEST_SUPP) "
1be5c7
+                            "error: %d", rc);
1be5c7
+            }
1be5c7
+            return ret;
1be5c7
         }
1be5c7
         ret = (reg == R_EAX) ? bitmask : bitmask >> 32;
1be5c7
     } else if (function == 0x80000001 && reg == R_ECX) {
1be5c7
-- 
1be5c7
2.35.3
1be5c7