Re: [PATCH 1/1] KVM: MMU: Fix VM entry failure and OOPS for shdaow page table

From: Sean Christopherson
Date: Tue Jun 07 2022 - 10:59:13 EST

Next message: Jan Beulich: "[PATCH] x86: drop bogus "cc" clobber from __try_cmpxchg_user_asm()"
Previous message: Rob Clark: "Re: [PATCH v2 1/2] drm: Add DRM_GEM_FOPS"
In reply to: Kai Huang: "Re: [PATCH 1/1] KVM: MMU: Fix VM entry failure and OOPS for shdaow page table"
Next in thread: Yuan Yao: "Re: [PATCH 1/1] KVM: MMU: Fix VM entry failure and OOPS for shdaow page table"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

"KVM: x86/mmu:" for the scope please (because I'm holding out hope that someday
we'll have a common "KVM: MMU:" that's shared by multiple architectures.

And there's a s/shdaow/shadow typo. That said, I'd prefer to use the shortlog to
give the reader a hint as to what the underlying bug is, e.g. it's not immediately
obvious that hitting this bug requires a platform with MKTME enabled. Maybe:

KVM: x86/mmu: Set memory encryption "value", not "mask", in shadow PDPTRs

On Tue, Jun 07, 2022, Yuan Yao wrote:
> commit e54f1ff244ac ("KVM: x86/mmu: Add shadow_me_value and

Personal preference, I don't bother with a "commit ..." in the changelog if there's
a Fixes: that provides the same information.

> repurpose shadow_me_mask") repurposed below varables:

Please wrap closer to ~75 chars.

Please lead with the "what", then dive into the details, that way the reader has
an idea of what's changing. And if there's a trace, finding the one sentence that
describes the actual change can be surprisingly difficult. Somethine like:

Assign shadow_me_value, not shadow_me_mask, to PAE root entries, a.k.a. shadow
PDPTRs, when host memory encryption is supported. The "mask" is the set of all
possible memory encryption bits, e.g. MKTME KeyIDs, whereas "value" holds the
actual value that needs to be stuffed into host page tables.

Using shadow_me_mask results in a failed VM-Entry due to setting reserved PA
bits in the PDPTRs, and ultimately causes an OOPS due to physical addresses
with non-zero MKTME bits sending to_shadow_page() into the weeds.

> shadow_me_value: the memory encryption bit(s) that will be
> set to the SPTE (the original shadow_me_mask).
> shadow_me_mask: all possible memory encryption bits (which
> is a super set of shadow_me_value).
>
> So assign shadow_me_mask to pae root page is wrong, instead
> using shadow_me_value.
>
> Fixes: e54f1ff244ac ("KVM: x86/mmu: Add shadow_me_value and repurpose shadow_me_mask")

Convention is to put Fixes: at the end of the changelog, i.e. after the trace and
next to the Cc list and SOB chain.

> ----------------------
> KVM: entry failed, hardware error 0x80000021

For this specific bug, I wouldn't bother with a dump of the failed VM-Entry,
nothing in the dump is relevant/interesting.

> If you're running a guest on an Intel machine without unrestricted mode
> support, the failure can be most likely due to the guest entering an invalid
> state for Intel VT. For example, the guest maybe running in big real mode
> which is not supported on less recent Intel processors.
>
> EAX=00000000 EBX=00000000 ECX=00000000 EDX=000806f3
> ESI=00000000 EDI=00000000 EBP=00000000 ESP=00000000
> EIP=0000e05b EFL=00000002 [-------] CPL=0 II=0 A20=1 SMM=0 HLT=0
> ES =0000 00000000 0000ffff 00009300
> CS =f000 000f0000 0000ffff 00009b00
> SS =0000 00000000 0000ffff 00009300
> DS =0000 00000000 0000ffff 00009300
> FS =0000 00000000 0000ffff 00009300
> GS =0000 00000000 0000ffff 00009300
> LDT=0000 00000000 0000ffff 00008200
> TR =0000 00000000 0000ffff 00008b00
> GDT= 00000000 0000ffff
> IDT= 00000000 0000ffff
> CR0=60000010 CR2=00000000 CR3=00000000 CR4=00000000
> DR0=0000000000000000 DR1=0000000000000000 DR2=0000000000000000 DR3=0000000000000000
> DR6=00000000ffff0ff0 DR7=0000000000000400
> EFER=0000000000000000
> Code=8c 0a 14 28 3c 50 64 c8 66 90 66 90 66 90 66 90 66 90 66 90 <2e> 66 83 3e c8 61 00 0f 85 89 f0 31 d2 8e d2 66 bc 00 70 00 00 66 ba 63 fc 0e 00 e9 f3 ee
>
> ----------------------
> [ 80.806596] set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.
> [ 293.504118] BUG: unable to handle page fault for address: ffd43f00063049e8
> [ 293.515075] #PF: supervisor read access in kernel mode
> [ 293.524031] #PF: error_code(0x0000) - not-present page
> [ 293.532935] PGD 86dfd8067 P4D 0
> [ 293.539626] Oops: 0000 [#1] PREEMPT SMP
> [ 293.546958] CPU: 164 PID: 4260 Comm: qemu-system-x86 Tainted: G W 5.18.0-rc6-kvm-upstream-workaround+ #82
> [ 293.565354] Hardware name: Intel Corporation ArcherCity/ArcherCity, BIOS EGSDCRB1.86B.0069.D14.2111291356 11/29/2021

Thanks for the trace! But please trim the superfluous information, e.g. the
timestamps, registers, code stream, etc... aren't necessary to understand why the
fault occured.

> [ 293.583639] RIP: 0010:mmu_free_root_page+0x3c/0x90 [kvm]
> [ 293.592911] Code: 25 28 00 00 00 48 89 45 f0 31 c0 48 8b 06 48 83 f8 ff 74 4a 48 c1 e0 0c 48 89 f3 48 c1 e8 18 48 c1 e0 06 48 03 05 e4 08 20 c2 <48> 8b 70 28 48 85 f6 74 41 80 7e 20 00 75 17 83 6e 48 01 75 18 f6
> [ 293.624056] RSP: 0018:ffa000000b3afb88 EFLAGS: 00010286
> [ 293.633326] RAX: ffd43f00063049c0 RBX: ff110000777ff000 RCX: 0000000000000001
> [ 293.644758] RDX: ffa000000b3afbc8 RSI: ff110000777ff000 RDI: ffa000000c211000
> [ 293.656132] RBP: ffa000000b3afba0 R08: 0000000000000100 R09: ffa000000b3afbe0
> [ 293.667480] R10: ffa000000b3afbe0 R11: 0000000000000000 R12: ffa000000c211000
> [ 293.678771] R13: ffa000000b3afbc8 R14: ff11000112e9c290 R15: 00000000ffffffef
> [ 293.690069] FS: 0000000000000000(0000) GS:ff1100084e300000(0000) knlGS:0000000000000000
> [ 293.702514] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [ 293.712338] CR2: ffd43f00063049e8 CR3: 000000000260a006 CR4: 0000000000773ee0
> [ 293.723802] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> [ 293.735230] DR3: 0000000000000000 DR6: 00000000fffe07f0 DR7: 0000000000000400
> [ 293.746597] PKRU: 55555554
> [ 293.752962] Call Trace:
> [ 293.758978] <TASK>
> [ 293.764579] kvm_mmu_free_roots+0xd1/0x200 [kvm]
> [ 293.773060] __kvm_mmu_unload+0x29/0x70 [kvm]
> [ 293.781177] kvm_mmu_unload+0x13/0x20 [kvm]
> [ 293.789012] kvm_arch_destroy_vm+0x8a/0x190 [kvm]
> [ 293.797355] kvm_put_kvm+0x197/0x2d0 [kvm]
> [ 293.804925] kvm_vm_release+0x21/0x30 [kvm]
> [ 293.812499] __fput+0x8e/0x260
> [ 293.818715] ____fput+0xe/0x10
> [ 293.824822] task_work_run+0x6f/0xb0
> [ 293.831433] do_exit+0x327/0xa90
> [ 293.837586] ? futex_unqueue+0x3f/0x70

Lines with leading '?' should be dropped, they're "guesses" from the unwinder.

> [ 293.844283] do_group_exit+0x35/0xa0
> [ 293.850770] get_signal+0x911/0x930
> [ 293.857137] arch_do_signal_or_restart+0x37/0x720
> [ 293.864896] ? do_futex+0xf9/0x1a0
> [ 293.871139] ? __x64_sys_futex+0x66/0x160
> [ 293.878001] exit_to_user_mode_prepare+0xb2/0x140
> [ 293.885576] syscall_exit_to_user_mode+0x16/0x30
> [ 293.892973] do_syscall_64+0x4e/0x90
> [ 293.899162] entry_SYSCALL_64_after_hwframe+0x44/0xae
> [ 293.906972] RIP: 0033:0x7f6c844f752d

Everything below here can be dropped as it's not relevant to the original bug.

E.g. the entire trace can be trimmed to:

set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.
BUG: unable to handle page fault for address: ffd43f00063049e8
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
PGD 86dfd8067 P4D 0
Oops: 0000 [#1] PREEMPT SMP
CPU: 164 PID: 4260 Comm: qemu-system-x86 Tainted: G W 5.18.0-rc6-kvm-upstream-workaround+ #82
Hardware name: Intel Corporation ArcherCity/ArcherCity, BIOS EGSDCRB1.86B.0069.D14.2111291356 11/29/2021
RIP: 0010:mmu_free_root_page+0x3c/0x90 [kvm]
Call Trace:
<TASK>
kvm_mmu_free_roots+0xd1/0x200 [kvm]
__kvm_mmu_unload+0x29/0x70 [kvm]
kvm_mmu_unload+0x13/0x20 [kvm]
kvm_arch_destroy_vm+0x8a/0x190 [kvm]
kvm_put_kvm+0x197/0x2d0 [kvm]
kvm_vm_release+0x21/0x30 [kvm]
__fput+0x8e/0x260
____fput+0xe/0x10
task_work_run+0x6f/0xb0
do_exit+0x327/0xa90
do_group_exit+0x35/0xa0
get_signal+0x911/0x930
arch_do_signal_or_restart+0x37/0x720
exit_to_user_mode_prepare+0xb2/0x140
syscall_exit_to_user_mode+0x16/0x30
do_syscall_64+0x4e/0x90
entry_SYSCALL_64_after_hwframe+0x44/0xae

> [ 293.913050] Code: Unable to access opcode bytes at RIP 0x7f6c844f7503.
> [ 293.922442] RSP: 002b:00007f6c7fbfe648 EFLAGS: 00000212 ORIG_RAX: 00000000000000ca
> [ 293.933048] RAX: fffffffffffffe00 RBX: 0000000000000000 RCX: 00007f6c844f752d
> [ 293.943161] RDX: 00000000ffffffff RSI: 0000000000000000 RDI: 0000557dd5a1ac58
> [ 293.953281] RBP: 00007f6c7fbfe670 R08: 0000000000000000 R09: 0000000000000000
> [ 293.963401] R10: 0000000000000000 R11: 0000000000000212 R12: 00007ffd0f3001be
> [ 293.973542] R13: 00007ffd0f3001bf R14: 00007ffd0f300280 R15: 00007f6c7fbfe880
> [ 293.983683] </TASK>
> [ 293.988216] Modules linked in: kvm_intel kvm x86_pkg_temp_thermal snd_pcm input_leds snd_timer joydev led_class snd irqbypass efi_pstore soundcore mac_hid button sch_fq_codel ip_tables x_tables ixgbe mdio mdio_devres libphy igc xfrm_algo ptp pps_core efivarfs [last unloaded: kvm]
> [ 294.022648] CR2: ffd43f00063049e8
> [ 294.028694] ---[ end trace 0000000000000000 ]---
> [ 294.042573] RIP: 0010:mmu_free_root_page+0x3c/0x90 [kvm]
> [ 294.050908] Code: 25 28 00 00 00 48 89 45 f0 31 c0 48 8b 06 48 83 f8 ff 74 4a 48 c1 e0 0c 48 89 f3 48 c1 e8 18 48 c1 e0 06 48 03 05 e4 08 20 c2 <48> 8b 70 28 48 85 f6 74 41 80 7e 20 00 75 17 83 6e 48 01 75 18 f6
> [ 294.079460] RSP: 0018:ffa000000b3afb88 EFLAGS: 00010286
> [ 294.087908] RAX: ffd43f00063049c0 RBX: ff110000777ff000 RCX: 0000000000000001
> [ 294.098558] RDX: ffa000000b3afbc8 RSI: ff110000777ff000 RDI: ffa000000c211000
> [ 294.109193] RBP: ffa000000b3afba0 R08: 0000000000000100 R09: ffa000000b3afbe0
> [ 294.119831] R10: ffa000000b3afbe0 R11: 0000000000000000 R12: ffa000000c211000
> [ 294.130449] R13: ffa000000b3afbc8 R14: ff11000112e9c290 R15: 00000000ffffffef
> [ 294.141090] FS: 0000000000000000(0000) GS:ff1100084e300000(0000) knlGS:0000000000000000
> [ 294.152867] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [ 294.162035] CR2: ffd43f00063049e8 CR3: 000000000260a006 CR4: 0000000000773ee0
> [ 294.172825] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> [ 294.183618] DR3: 0000000000000000 DR6: 00000000fffe07f0 DR7: 0000000000000400
> [ 294.194390] PKRU: 55555554
> [ 294.200189] note: qemu-system-x86[4260] exited with preempt_count 1
> [ 294.210044] Fixing recursive fault but reboot is needed!
> [ 294.218854] BUG: scheduling while atomic: qemu-system-x86/4260/0x00000000
> [ 294.229357] Modules linked in: kvm_intel kvm x86_pkg_temp_thermal snd_pcm input_leds snd_timer joydev led_class snd irqbypass efi_pstore soundcore mac_hid button sch_fq_codel ip_tables x_tables ixgbe mdio mdio_devres libphy igc xfrm_algo ptp pps_core efivarfs [last unloaded: kvm]
> [ 294.266273] Preemption disabled at:
> [ 294.266273] [<ffffffff8109e404>] do_task_dead+0x24/0x50
> [ 294.282274] CPU: 164 PID: 4260 Comm: qemu-system-x86 Tainted: G D W 5.18.0-rc6-kvm-upstream-workaround+ #82
> [ 294.300693] Hardware name: Intel Corporation ArcherCity/ArcherCity, BIOS EGSDCRB1.86B.0069.D14.2111291356 11/29/2021
> [ 294.319002] Call Trace:
> [ 294.325007] <TASK>
> [ 294.330587] dump_stack_lvl+0x38/0x49
> [ 294.337889] ? do_task_dead+0x24/0x50
> [ 294.345102] dump_stack+0x10/0x12
> [ 294.351836] __schedule_bug.cold.156+0x7d/0x8e
> [ 294.359770] __schedule+0x578/0x820
> [ 294.366552] ? vprintk+0x52/0x80
> [ 294.373025] ? _printk+0x58/0x6f
> [ 294.379449] do_task_dead+0x44/0x50
> [ 294.386097] make_task_dead.cold.48+0x50/0xaf
> [ 294.393650] rewind_stack_and_make_dead+0x17/0x17
> [ 294.401549] RIP: 0033:0x7f6c844f752d
> [ 294.408147] Code: Unable to access opcode bytes at RIP 0x7f6c844f7503.
> [ 294.418086] RSP: 002b:00007f6c7fbfe648 EFLAGS: 00000212 ORIG_RAX: 00000000000000ca
> [ 294.429266] RAX: fffffffffffffe00 RBX: 0000000000000000 RCX: 00007f6c844f752d
> [ 294.439998] RDX: 00000000ffffffff RSI: 0000000000000000 RDI: 0000557dd5a1ac58
> [ 294.450748] RBP: 00007f6c7fbfe670 R08: 0000000000000000 R09: 0000000000000000
> [ 294.461498] R10: 0000000000000000 R11: 0000000000000212 R12: 00007ffd0f3001be
> [ 294.472199] R13: 00007ffd0f3001bf R14: 00007ffd0f300280 R15: 00007f6c7fbfe880
> [ 294.482869] </TASK>
>
> Signed-off-by: Yuan Yao <yuan.yao@xxxxxxxxx>
> ---
> arch/x86/kvm/mmu/mmu.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> index efe5a3dca1e0..6bd144f1e60c 100644
> --- a/arch/x86/kvm/mmu/mmu.c
> +++ b/arch/x86/kvm/mmu/mmu.c
> @@ -3411,7 +3411,7 @@ static int mmu_alloc_direct_roots(struct kvm_vcpu *vcpu)
> root = mmu_alloc_root(vcpu, i << (30 - PAGE_SHIFT),
> i << 30, PT32_ROOT_LEVEL, true);
> mmu->pae_root[i] = root | PT_PRESENT_MASK |
> - shadow_me_mask;
> + shadow_me_value;
> }
> mmu->root.hpa = __pa(mmu->pae_root);
> } else {
> --
> 2.27.0
>

Next message: Jan Beulich: "[PATCH] x86: drop bogus "cc" clobber from __try_cmpxchg_user_asm()"
Previous message: Rob Clark: "Re: [PATCH v2 1/2] drm: Add DRM_GEM_FOPS"
In reply to: Kai Huang: "Re: [PATCH 1/1] KVM: MMU: Fix VM entry failure and OOPS for shdaow page table"
Next in thread: Yuan Yao: "Re: [PATCH 1/1] KVM: MMU: Fix VM entry failure and OOPS for shdaow page table"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]