Skip to contentSkip to conversation

WARNING in kvm_read_guest_offset_cached

syzbot Superseded syzbot 5 participants 8 messages Jul 16, 2025 – Apr 6

Subsystems (3)
lore
Share & tools
lore

syzbot

author syzbot
#0 lore raw

Hello,

syzbot found the following issue on:

HEAD commit:    155a3c003e55 Merge tag 'for-6.16/dm-fixes-2' of git://git...
git tree:       upstream
console output: https://syzkaller.appspot.com/x/log.txt?x=103e858c580000
kernel config:  https://syzkaller.appspot.com/x/.config?x=8d5ef2da1e1c848
dashboard link: https://syzkaller.appspot.com/bug?extid=bc0e18379a290e5edfe4
compiler:       gcc (Debian 12.2.0-14) 12.2.0, GNU ld (GNU Binutils for Debian) 2.40
syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=153188f0580000
C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=16f6198c580000

Downloadable assets:
disk image (non-bootable): https://storage.googleapis.com/syzbot-assets/d900f083ada3/non_bootable_disk-155a3c00.raw.xz
vmlinux: https://storage.googleapis.com/syzbot-assets/725a320dfe66/vmlinux-155a3c00.xz
kernel image: https://storage.googleapis.com/syzbot-assets/9f06899bb6f3/bzImage-155a3c00.xz

IMPORTANT: if you fix the issue, please add the following tag to the commit:

Reported-by [email protected]
------------[ cut here ]------------
WARNING: CPU: 0 PID: 6107 at arch/x86/kvm/../../../virt/kvm/kvm_main.c:3459 kvm_read_guest_offset_cached+0x3f5/0x4b0 virt/kvm/kvm_main.c:3459
Modules linked in:
CPU: 0 UID: 0 PID: 6107 Comm: syz.0.16 Not tainted 6.16.0-rc6-syzkaller-00002-g155a3c003e55 #0 PREEMPT(full) 
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.3-debian-1.16.3-2~bpo12+1 04/01/2014
RIP: 0010:kvm_read_guest_offset_cached+0x3f5/0x4b0 virt/kvm/kvm_main.c:3459
Code: 0f 01 e8 3e 6c 61 00 e9 9b fc ff ff e8 14 25 85 00 48 8b 3c 24 31 d2 48 89 ee e8 16 bf fa 00 e9 2e fe ff ff e8 fc 24 85 00 90 <0f> 0b 90 bb ea ff ff ff e9 4d fe ff ff e8 e9 24 85 00 48 8b 74 24
RSP: 0018:ffffc9000349f960 EFLAGS: 00010293
RAX: 0000000000000000 RBX: ffff888050329898 RCX: ffffffff8136ca66
RDX: ffff88803cfa8000 RSI: ffffffff8136cd84 RDI: 0000000000000006
RBP: 0000000000000004 R08: 0000000000000006 R09: 0000000000000008
R10: 0000000000000000 R11: 0000000000000001 R12: 0000000000000004
R13: ffffc90003921000 R14: 0000000000000000 R15: ffffc900039215a0
FS:  000055558378f500(0000) GS:ffff8880d6713000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 0000000025de6000 CR4: 0000000000352ef0
Call Trace:
 <TASK>
 apf_pageready_slot_free arch/x86/kvm/x86.c:13452 [inline]
 kvm_arch_can_dequeue_async_page_present+0x172/0x310 arch/x86/kvm/x86.c:13571
 kvm_check_async_pf_completion+0x61/0x440 virt/kvm/async_pf.c:158
 vcpu_enter_guest arch/x86/kvm/x86.c:10930 [inline]
 vcpu_run+0x48c2/0x5500 arch/x86/kvm/x86.c:11335
 kvm_arch_vcpu_ioctl_run+0x51e/0x18c0 arch/x86/kvm/x86.c:11658
 kvm_vcpu_ioctl+0x5eb/0x1690 virt/kvm/kvm_main.c:4464
 vfs_ioctl fs/ioctl.c:51 [inline]
 __do_sys_ioctl fs/ioctl.c:907 [inline]
 __se_sys_ioctl fs/ioctl.c:893 [inline]
 __x64_sys_ioctl+0x18e/0x210 fs/ioctl.c:893
 do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
 do_syscall_64+0xcd/0x4c0 arch/x86/entry/syscall_64.c:94
 entry_SYSCALL_64_after_hwframe+0x77/0x7f
RIP: 0033:0x7fdb7658e929
Code: ff ff c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 40 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 c7 c1 a8 ff ff ff f7 d8 64 89 01 48
RSP: 002b:00007ffd25fe7d98 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
RAX: ffffffffffffffda RBX: 00007fdb767b5fa0 RCX: 00007fdb7658e929
RDX: 0000000000000000 RSI: 000000000000ae80 RDI: 0000000000000005
RBP: 00007fdb76610b39 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000
R13: 00007fdb767b5fa0 R14: 00007fdb767b5fa0 R15: 0000000000000003
 </TASK>

---
This report is generated by a bot. It may contain errors. See https://goo.gl/tpsmEJ for more information about syzbot. syzbot engineers can be reached at [email protected].

syzbot will keep track of this issue. See:
https://goo.gl/tpsmEJ#status for how to communicate with syzbot.

If the report is already addressed, let syzbot know by replying with:
#syz fix: exact-commit-title

If you want syzbot to run the reproducer, reply with:
#syz test: git://repo/address.git branch-or-commit-hash
If you attach or paste a git patch, syzbot will apply it before testing.

If you want to overwrite report's subsystems, reply with:
#syz set subsystems: new-subsystem
(See the list of subsystem names on the web dashboard)

If the report is a duplicate of another one, reply with:
#syz dup: exact-subject-of-another-report

If you want to undo deduplication, reply with:
#syz undup

#1 lore raw
Quoted 32 lines · syzbot

Hello,

syzbot found the following issue on:

HEAD commit:    155a3c003e55 Merge tag 'for-6.16/dm-fixes-2' of git://git...
git tree:       upstream
console output: https://syzkaller.appspot.com/x/log.txt?x=103e858c580000
kernel config:  https://syzkaller.appspot.com/x/.config?x=8d5ef2da1e1c848
dashboard link: https://syzkaller.appspot.com/bug?extid=bc0e18379a290e5edfe4
compiler:       gcc (Debian 12.2.0-14) 12.2.0, GNU ld (GNU Binutils for Debian) 2.40
syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=153188f0580000
C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=16f6198c580000

Downloadable assets:
disk image (non-bootable): https://storage.googleapis.com/syzbot-assets/d900f083ada3/non_bootable_disk-155a3c00.raw.xz
vmlinux: https://storage.googleapis.com/syzbot-assets/725a320dfe66/vmlinux-155a3c00.xz
kernel image: https://storage.googleapis.com/syzbot-assets/9f06899bb6f3/bzImage-155a3c00.xz

IMPORTANT: if you fix the issue, please add the following tag to the commit:

Reported-by [email protected]
------------[ cut here ]------------
WARNING: CPU: 0 PID: 6107 at arch/x86/kvm/../../../virt/kvm/kvm_main.c:3459 kvm_read_guest_offset_cached+0x3f5/0x4b0 virt/kvm/kvm_main.c:3459
Modules linked in:
CPU: 0 UID: 0 PID: 6107 Comm: syz.0.16 Not tainted 6.16.0-rc6-syzkaller-00002-g155a3c003e55 #0 PREEMPT(full) 
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.3-debian-1.16.3-2~bpo12+1 04/01/2014
RIP: 0010:kvm_read_guest_offset_cached+0x3f5/0x4b0 virt/kvm/kvm_main.c:3459
Code: 0f 01 e8 3e 6c 61 00 e9 9b fc ff ff e8 14 25 85 00 48 8b 3c 24 31 d2 48 89 ee e8 16 bf fa 00 e9 2e fe ff ff e8 fc 24 85 00 90 <0f> 0b 90 bb ea ff ff ff e9 4d fe ff ff e8 e9 24 85 00 48 8b 74 24
RSP: 0018:ffffc9000349f960 EFLAGS: 00010293
RAX: 0000000000000000 RBX: ffff888050329898 RCX: ffffffff8136ca66
RDX: ffff88803cfa8000 RSI: ffffffff8136cd84 RDI: 0000000000000006
RBP: 0000000000000004 R08: 0000000000000006 R09: 0000000000000008
Quoted 8 lines
R10: 0000000000000000 R11: 0000000000000001 R12: 0000000000000004
R13: ffffc90003921000 R14: 0000000000000000 R15: ffffc900039215a0
FS:  000055558378f500(0000) GS:ffff8880d6713000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 0000000025de6000 CR4: 0000000000352ef0
Call Trace:
 <TASK>
 apf_pageready_slot_free arch/x86/kvm/x86.c:13452 [inline]

kvm_pv_enable_async_pf() sets vcpu->arch.apf.msr_en_val even if the gpa is bad, which leaves the cache in an empty state. Something like so over a few patches fixes the problem:

---
 arch/x86/kvm/x86.c | 32 ++++++++++++++++++--------------
 1 file changed, 18 insertions(+), 14 deletions(-)
arch/x86/kvm/x86.c +18 -14
index a1c49bc681c4..0fbbf297b3c8 100644
@@ -3547,11 +3547,16 @@ static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 	return 0;
 }
 
-static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
+static bool __kvm_pv_async_pf_enabled(u64 msr_en_val)
 {
 	u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
 
-	return (vcpu->arch.apf.msr_en_val & mask) == mask;
+	return (msr_en_val & mask) == mask;
+}
+
+static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
+{
+	return __kvm_pv_async_pf_enabled(vcpu->arch.apf.msr_en_val);
 }
 
 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
@@ -3573,22 +3578,21 @@ static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
 	if (!lapic_in_kernel(vcpu))
 		return data ? 1 : 0;
 
+	if (__kvm_pv_async_pf_enabled(data) &&
+	    kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
+				      sizeof(u64)))
+		return 1;
+
 	vcpu->arch.apf.msr_en_val = data;
-
-	if (!kvm_pv_async_pf_enabled(vcpu)) {
-		kvm_clear_async_pf_completion_queue(vcpu);
-		kvm_async_pf_hash_reset(vcpu);
-		return 0;
-	}
-
-	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
-					sizeof(u64)))
-		return 1;
-
 	vcpu->arch.apf.send_always = (data & KVM_ASYNC_PF_SEND_ALWAYS);
 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
 
-	kvm_async_pf_wakeup_all(vcpu);
+	if (kvm_pv_async_pf_enabled(vcpu)) {
+		kvm_clear_async_pf_completion_queue(vcpu);
+		kvm_async_pf_hash_reset(vcpu);
+	} else {
+		kvm_async_pf_wakeup_all(vcpu);
+	}
 
 	return 0;
 }

base-commit: 2a046f6a4ecce47ada50dba529ec726dd0d34351 --

#2 lore raw
Quoted 43 lines · Sean Christopherson' via syzkaller-bugs
Quoted 32 lines · syzbot

Hello,

syzbot found the following issue on:

HEAD commit:    155a3c003e55 Merge tag 'for-6.16/dm-fixes-2' of git://git...
git tree:       upstream
console output: https://syzkaller.appspot.com/x/log.txt?x=103e858c580000
kernel config:  https://syzkaller.appspot.com/x/.config?x=8d5ef2da1e1c848
dashboard link: https://syzkaller.appspot.com/bug?extid=bc0e18379a290e5edfe4
compiler:       gcc (Debian 12.2.0-14) 12.2.0, GNU ld (GNU Binutils for Debian) 2.40
syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=153188f0580000
C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=16f6198c580000

Downloadable assets:
disk image (non-bootable): https://storage.googleapis.com/syzbot-assets/d900f083ada3/non_bootable_disk-155a3c00.raw.xz
vmlinux: https://storage.googleapis.com/syzbot-assets/725a320dfe66/vmlinux-155a3c00.xz
kernel image: https://storage.googleapis.com/syzbot-assets/9f06899bb6f3/bzImage-155a3c00.xz

IMPORTANT: if you fix the issue, please add the following tag to the commit:

Reported-by [email protected]
------------[ cut here ]------------
WARNING: CPU: 0 PID: 6107 at arch/x86/kvm/../../../virt/kvm/kvm_main.c:3459 kvm_read_guest_offset_cached+0x3f5/0x4b0 virt/kvm/kvm_main.c:3459
Modules linked in:
CPU: 0 UID: 0 PID: 6107 Comm: syz.0.16 Not tainted 6.16.0-rc6-syzkaller-00002-g155a3c003e55 #0 PREEMPT(full)
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.3-debian-1.16.3-2~bpo12+1 04/01/2014
RIP: 0010:kvm_read_guest_offset_cached+0x3f5/0x4b0 virt/kvm/kvm_main.c:3459
Code: 0f 01 e8 3e 6c 61 00 e9 9b fc ff ff e8 14 25 85 00 48 8b 3c 24 31 d2 48 89 ee e8 16 bf fa 00 e9 2e fe ff ff e8 fc 24 85 00 90 <0f> 0b 90 bb ea ff ff ff e9 4d fe ff ff e8 e9 24 85 00 48 8b 74 24
RSP: 0018:ffffc9000349f960 EFLAGS: 00010293
RAX: 0000000000000000 RBX: ffff888050329898 RCX: ffffffff8136ca66
RDX: ffff88803cfa8000 RSI: ffffffff8136cd84 RDI: 0000000000000006
RBP: 0000000000000004 R08: 0000000000000006 R09: 0000000000000008
Quoted 8 lines
R10: 0000000000000000 R11: 0000000000000001 R12: 0000000000000004
R13: ffffc90003921000 R14: 0000000000000000 R15: ffffc900039215a0
FS:  000055558378f500(0000) GS:ffff8880d6713000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 0000000025de6000 CR4: 0000000000352ef0
Call Trace:
 <TASK>
 apf_pageready_slot_free arch/x86/kvm/x86.c:13452 [inline]
Quoted 3 lines

kvm_pv_enable_async_pf() sets vcpu->arch.apf.msr_en_val even if the gpa is bad, which leaves the cache in an empty state. Something like so over a few patches fixes the problem:

This bug is still occasionally reproducible on an Intel host running 6.19.

#3 lore raw
Quoted 45 lines · Alexander Potapenko
Quoted 43 lines · Sean Christopherson' via syzkaller-bugs
Quoted 40 lines · syzbot

Hello,

syzbot found the following issue on:

HEAD commit:    155a3c003e55 Merge tag 'for-6.16/dm-fixes-2' of git://git...
git tree:       upstream
console output: https://syzkaller.appspot.com/x/log.txt?x=103e858c580000
kernel config:  https://syzkaller.appspot.com/x/.config?x=8d5ef2da1e1c848
dashboard link: https://syzkaller.appspot.com/bug?extid=bc0e18379a290e5edfe4
compiler:       gcc (Debian 12.2.0-14) 12.2.0, GNU ld (GNU Binutils for Debian) 2.40
syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=153188f0580000
C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=16f6198c580000

Downloadable assets:
disk image (non-bootable): https://storage.googleapis.com/syzbot-assets/d900f083ada3/non_bootable_disk-155a3c00.raw.xz
vmlinux: https://storage.googleapis.com/syzbot-assets/725a320dfe66/vmlinux-155a3c00.xz
kernel image: https://storage.googleapis.com/syzbot-assets/9f06899bb6f3/bzImage-155a3c00.xz

IMPORTANT: if you fix the issue, please add the following tag to the commit:

Reported-by [email protected]
------------[ cut here ]------------
WARNING: CPU: 0 PID: 6107 at arch/x86/kvm/../../../virt/kvm/kvm_main.c:3459 kvm_read_guest_offset_cached+0x3f5/0x4b0 virt/kvm/kvm_main.c:3459
Modules linked in:
CPU: 0 UID: 0 PID: 6107 Comm: syz.0.16 Not tainted 6.16.0-rc6-syzkaller-00002-g155a3c003e55 #0 PREEMPT(full)
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.3-debian-1.16.3-2~bpo12+1 04/01/2014
RIP: 0010:kvm_read_guest_offset_cached+0x3f5/0x4b0 virt/kvm/kvm_main.c:3459
Code: 0f 01 e8 3e 6c 61 00 e9 9b fc ff ff e8 14 25 85 00 48 8b 3c 24 31 d2 48 89 ee e8 16 bf fa 00 e9 2e fe ff ff e8 fc 24 85 00 90 <0f> 0b 90 bb ea ff ff ff e9 4d fe ff ff e8 e9 24 85 00 48 8b 74 24
RSP: 0018:ffffc9000349f960 EFLAGS: 00010293
RAX: 0000000000000000 RBX: ffff888050329898 RCX: ffffffff8136ca66
RDX: ffff88803cfa8000 RSI: ffffffff8136cd84 RDI: 0000000000000006
RBP: 0000000000000004 R08: 0000000000000006 R09: 0000000000000008
R10: 0000000000000000 R11: 0000000000000001 R12: 0000000000000004
R13: ffffc90003921000 R14: 0000000000000000 R15: ffffc900039215a0
FS:  000055558378f500(0000) GS:ffff8880d6713000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 0000000025de6000 CR4: 0000000000352ef0
Call Trace:
 <TASK>
 apf_pageready_slot_free arch/x86/kvm/x86.c:13452 [inline]
Quoted 5 lines
Quoted 3 lines

kvm_pv_enable_async_pf() sets vcpu->arch.apf.msr_en_val even if the gpa is bad, which leaves the cache in an empty state. Something like so over a few patches fixes the problem:

This bug is still occasionally reproducible on an Intel host running 6.19.

Gah, I never actually posted a series to fix this. I'll try to get that done this week.

#4 lore raw

PATCH 1/1 KVM: x86: Don't leave APF half-enabled on bad APF data GPA

In reply to syzbot

kvm_pv_enable_async_pf() updates vcpu->arch.apf.msr_en_val before initializing the APF data gfn_to_hva cache. If userspace provides an invalid GPA, kvm_gfn_to_hva_cache_init() fails, but msr_en_val stays enabled and leaves APF state half-initialized.

Later APF paths can then try to use the empty cache and trigger WARN_ON() in kvm_read_guest_offset_cached().

Determine the new APF enabled state from the incoming MSR value, do cache initialization first on the enable path, and commit msr_en_val only after successful initialization. Keep the disable path behavior unchanged.

---
 arch/x86/kvm/x86.c | 19 +++++++++++++------
 1 file changed, 13 insertions(+), 6 deletions(-)
arch/x86/kvm/x86.c +13 -6
index fd1c4a36b59..c355bbd36c3 100644
@@ -1041,11 +1041,15 @@ bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
 }
 EXPORT_SYMBOL_FOR_KVM_INTERNAL(kvm_require_dr);
 
-static bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
+static bool kvm_pv_async_pf_enabled_data(u64 data)
 {
 	u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
+	return (data & mask) == mask;
+}
 
-	return (vcpu->arch.apf.msr_en_val & mask) == mask;
+static bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
+{
+	return kvm_pv_async_pf_enabled_data(vcpu->arch.apf.msr_en_val);
 }
 
 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
@@ -3645,17 +3649,20 @@ static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
 	if (!lapic_in_kernel(vcpu))
 		return data ? 1 : 0;
 
+	bool new_enabled = kvm_pv_async_pf_enabled_data(data);
+
+	if (new_enabled &&
+		kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
+			sizeof(u64)))
+		return 1;
 	vcpu->arch.apf.msr_en_val = data;
 
-	if (!kvm_pv_async_pf_enabled(vcpu)) {
+	if (!new_enabled) {
 		kvm_clear_async_pf_completion_queue(vcpu);
 		kvm_async_pf_hash_reset(vcpu);
 		return 0;
 	}
 
-	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
-					sizeof(u64)))
-		return 1;
 
 	vcpu->arch.apf.send_always = (data & KVM_ASYNC_PF_SEND_ALWAYS);
 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
Signature · 1 line
2.47.3
Reviewed-by on v1 1/1
#5 lore raw

Re: [PATCH] KVM: x86: Don't leave APF half-enabled on bad APF data GPA

Quoted 12 lines · Ethan Yang

kvm_pv_enable_async_pf() updates vcpu->arch.apf.msr_en_val before initializing the APF data gfn_to_hva cache. If userspace provides an invalid GPA, kvm_gfn_to_hva_cache_init() fails, but msr_en_val stays enabled and leaves APF state half-initialized.

Later APF paths can then try to use the empty cache and trigger WARN_ON() in kvm_read_guest_offset_cached().

Determine the new APF enabled state from the incoming MSR value, do cache initialization first on the enable path, and commit msr_en_val only after successful initialization. Keep the disable path behavior unchanged.

Quoted 5 lines

though some nits below.

Quoted 9 lines
---
  arch/x86/kvm/x86.c | 19 +++++++++++++------
  1 file changed, 13 insertions(+), 6 deletions(-)
arch/x86/kvm/x86.c +0 -0
index fd1c4a36b59..c355bbd36c3 100644

arch/x86/kvm/x86.c bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr) · line 1044

Quoted 5 lines
arch/x86/kvm/x86.c +1 -1
@@ -1041,11 +1041,15 @@ bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
  }
  EXPORT_SYMBOL_FOR_KVM_INTERNAL(kvm_require_dr);
 
-static bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
+static bool kvm_pv_async_pf_enabled_data(u64 data)

I'm not sure if the name with "__" as __kvm_pv_async_pf_enabled() fits better?

Quoted 13 lines
arch/x86/kvm/x86.c +5 -1
  {
  	u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
+	return (data & mask) == mask;
+}
 
-	return (vcpu->arch.apf.msr_en_val & mask) == mask;
+static bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
+{
+	return kvm_pv_async_pf_enabled_data(vcpu->arch.apf.msr_en_val);
  }
 
  static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)

arch/x86/kvm/x86.c static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data) · line 3652

Quoted 4 lines
arch/x86/kvm/x86.c +1 -0
@@ -3645,17 +3649,20 @@ static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
  	if (!lapic_in_kernel(vcpu))
  		return data ? 1 : 0;
 
+	bool new_enabled = kvm_pv_async_pf_enabled_data(data);

I would like just call it "enable"

arch/x86/kvm/x86.c static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data) · line 3655

Quoted 3 lines
arch/x86/kvm/x86.c +3 -0
+
+	if (new_enabled &&
+		kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,

Please align the second line

Quoted 17 lines
arch/x86/kvm/x86.c +3 -4
+			sizeof(u64)))
+		return 1;
  	vcpu->arch.apf.msr_en_val = data;
 
-	if (!kvm_pv_async_pf_enabled(vcpu)) {
+	if (!new_enabled) {
  		kvm_clear_async_pf_completion_queue(vcpu);
  		kvm_async_pf_hash_reset(vcpu);
  		return 0;
  	}
 
-	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
-					sizeof(u64)))
-		return 1;
 
  	vcpu->arch.apf.send_always = (data & KVM_ASYNC_PF_SEND_ALWAYS);
  	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
#6 lore raw

PATCH v2 1/1 KVM: x86: Don't leave APF half-enabled on bad APF data GPA

In reply to Ethan Yang

kvm_pv_enable_async_pf() updates vcpu->arch.apf.msr_en_val before initializing the APF data gfn_to_hva cache. If userspace provides an invalid GPA, kvm_gfn_to_hva_cache_init() fails, but msr_en_val stays enabled and leaves APF state half-initialized.

Later APF paths can then try to use the empty cache and trigger WARN_ON() in kvm_read_guest_offset_cached().

Determine the new APF enabled state from the incoming MSR value, do cache initialization first on the enable path, and commit msr_en_val only after successful initialization. Keep the disable path behavior unchanged.

---
v2:
- Rename helper to __kvm_pv_async_pf_enabled()
- Rename new_enabled to enable
- Align wrapped line in kvm_gfn_to_hva_cache_init() call

---
 arch/x86/kvm/x86.c | 22 +++++++++++++++-------
 1 file changed, 15 insertions(+), 7 deletions(-)
arch/x86/kvm/x86.c +15 -7
index 63afdb6bb07..93b7ad9b52b 100644
@@ -1035,11 +1035,16 @@ bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
 }
 EXPORT_SYMBOL_FOR_KVM_INTERNAL(kvm_require_dr);
 
-static bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
+static bool __kvm_pv_async_pf_enabled(u64 data)
 {
 	u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
 
-	return (vcpu->arch.apf.msr_en_val & mask) == mask;
+	return (data & mask) == mask;
+}
+
+static bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
+{
+	return __kvm_pv_async_pf_enabled(vcpu->arch.apf.msr_en_val);
 }
 
 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
@@ -3616,6 +3621,7 @@ static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
 {
 	gpa_t gpa = data & ~0x3f;
+	bool enable;
 
 	/* Bits 4:5 are reserved, Should be zero */
 	if (data & 0x30)
@@ -3632,18 +3638,20 @@ static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
 	if (!lapic_in_kernel(vcpu))
 		return data ? 1 : 0;
 
+	enable = __kvm_pv_async_pf_enabled(data);
+
+	if (enable &&
+	    kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
+				      sizeof(u64)))
+		return 1;
 	vcpu->arch.apf.msr_en_val = data;
 
-	if (!kvm_pv_async_pf_enabled(vcpu)) {
+	if (!enable) {
 		kvm_clear_async_pf_completion_queue(vcpu);
 		kvm_async_pf_hash_reset(vcpu);
 		return 0;
 	}
 
-	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
-					sizeof(u64)))
-		return 1;
-
 	vcpu->arch.apf.send_always = (data & KVM_ASYNC_PF_SEND_ALWAYS);
 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
 
Signature · 1 line
2.47.3
#7 lore raw

Re: [PATCH v2] KVM: x86: Don't leave APF half-enabled on bad APF data GPA

Thanks for posting this! My "week" estimate was wee bit off...

Quoted 10 lines · Ethan Yang

static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)

arch/x86/kvm/x86.c +1 -0
@@ -3616,6 +3621,7 @@ static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
 {
 	gpa_t gpa = data & ~0x3f;
+	bool enable;
 
 	/* Bits 4:5 are reserved, Should be zero */
 	if (data & 0x30)

arch/x86/kvm/x86.c static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data) · line 3645

Quoted 8 lines
arch/x86/kvm/x86.c +5 -0
@@ -3632,18 +3638,20 @@ static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
 	if (!lapic_in_kernel(vcpu))
 		return data ? 1 : 0;
 
+	enable = __kvm_pv_async_pf_enabled(data);
+
+	if (enable &&
+	    kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
+				      sizeof(u64)))

I would rather forgo a local variable and either hhave the below check stay as kvm_pv_async_pf_enabled() or just redo the call to __kvm_pv_async_pf_enabled().

arch/x86/kvm/x86.c static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data) · line 3646

Quoted 1 line
arch/x86/kvm/x86.c +1 -0
+		return 1;

Newline please.

Quoted 5 lines
arch/x86/kvm/x86.c +1 -1
 	vcpu->arch.apf.msr_en_val = data;
 
-	if (!kvm_pv_async_pf_enabled(vcpu)) {
+	if (!enable) {
 		kvm_clear_async_pf_completion_queue(vcpu);

arch/x86/kvm/x86.c static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data) · line 3656

Quoted 10 lines
arch/x86/kvm/x86.c +0 -4
 		kvm_async_pf_hash_reset(vcpu);
 		return 0;
 	}
 
-	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
-					sizeof(u64)))
-		return 1;
-
 	vcpu->arch.apf.send_always = (data & KVM_ASYNC_PF_SEND_ALWAYS);
 	vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;

As I sketched out, in a follow-up patch, I would like to to update these fields as well. I don't like tracking stale information, even if it _should_ be unused.

Actually, even better, just drop the fields. That way zeroing msr_en_val via INIT won't lead to stale data either.

I'll post a v3, should be easier overall than posting diffs for the suggestions and then making you write changelogs :-)