patches/kernel/0015-KVM-x86-emulator-smm-preserve-interrupt-shadow-in-SM.patch

   1 From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
   2 From: Maxim Levitsky <mlevitsk@redhat.com>
   3 Date: Wed, 3 Aug 2022 18:50:11 +0300
   4 Subject: [PATCH] KVM: x86: emulator/smm: preserve interrupt shadow in SMRAM
   5 MIME-Version: 1.0
   6 Content-Type: text/plain; charset=UTF-8
   7 Content-Transfer-Encoding: 8bit
   8
   9 When #SMI is asserted, the CPU can be in interrupt shadow
  10 due to sti or mov ss.
  11
  12 It is not mandatory in  Intel/AMD prm to have the #SMI
  13 blocked during the shadow, and on top of
  14 that, since neither SVM nor VMX has true support for SMI
  15 window, waiting for one instruction would mean single stepping
  16 the guest.
  17
  18 Instead, allow #SMI in this case, but both reset the interrupt
  19 window and stash its value in SMRAM to restore it on exit
  20 from SMM.
  21
  22 This fixes rare failures seen mostly on windows guests on VMX,
  23 when #SMI falls on the sti instruction which mainfest in
  24 VM entry failure due to EFLAGS.IF not being set, but STI interrupt
  25 window still being set in the VMCS.
  26
  27 Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
  28 Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
  29 Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
  30 ---
  31  arch/x86/kvm/emulate.c     | 17 ++++++++++++++---
  32  arch/x86/kvm/kvm_emulate.h | 10 ++++++----
  33  arch/x86/kvm/x86.c         | 12 ++++++++++++
  34  3 files changed, 32 insertions(+), 7 deletions(-)
  35
  36 diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
  37 index 03f9e5aa036e..bb008a5be539 100644
  38 --- a/arch/x86/kvm/emulate.c
  39 +++ b/arch/x86/kvm/emulate.c
  40 @@ -2435,7 +2435,7 @@ static int rsm_load_state_32(struct x86_emulate_ctxt *ctxt,
  41                              const struct kvm_smram_state_32 *smstate)
  42  {
  43         struct desc_ptr dt;
  44 -       int i;
  45 +       int i, r;
  46
  47         ctxt->eflags =  smstate->eflags | X86_EFLAGS_FIXED;
  48         ctxt->_eip =  smstate->eip;
  49 @@ -2470,8 +2470,16 @@ static int rsm_load_state_32(struct x86_emulate_ctxt *ctxt,
  50
  51         ctxt->ops->set_smbase(ctxt, smstate->smbase);
  52
  53 -       return rsm_enter_protected_mode(ctxt, smstate->cr0,
  54 -                                       smstate->cr3, smstate->cr4);
  55 +       r = rsm_enter_protected_mode(ctxt, smstate->cr0,
  56 +                                    smstate->cr3, smstate->cr4);
  57 +
  58 +       if (r != X86EMUL_CONTINUE)
  59 +               return r;
  60 +
  61 +       ctxt->ops->set_int_shadow(ctxt, 0);
  62 +       ctxt->interruptibility = (u8)smstate->int_shadow;
  63 +
  64 +       return X86EMUL_CONTINUE;
  65  }
  66
  67  #ifdef CONFIG_X86_64
  68 @@ -2520,6 +2528,9 @@ static int rsm_load_state_64(struct x86_emulate_ctxt *ctxt,
  69         rsm_load_seg_64(ctxt, &smstate->fs, VCPU_SREG_FS);
  70         rsm_load_seg_64(ctxt, &smstate->gs, VCPU_SREG_GS);
  71
  72 +       ctxt->ops->set_int_shadow(ctxt, 0);
  73 +       ctxt->interruptibility = (u8)smstate->int_shadow;
  74 +
  75         return X86EMUL_CONTINUE;
  76  }
  77  #endif
  78 diff --git a/arch/x86/kvm/kvm_emulate.h b/arch/x86/kvm/kvm_emulate.h
  79 index 76c0b8e7890b..a7313add0f2a 100644
  80 --- a/arch/x86/kvm/kvm_emulate.h
  81 +++ b/arch/x86/kvm/kvm_emulate.h
  82 @@ -234,6 +234,7 @@ struct x86_emulate_ops {
  83         bool (*guest_has_rdpid)(struct x86_emulate_ctxt *ctxt);
  84
  85         void (*set_nmi_mask)(struct x86_emulate_ctxt *ctxt, bool masked);
  86 +       void (*set_int_shadow)(struct x86_emulate_ctxt *ctxt, u8 shadow);
  87
  88         unsigned (*get_hflags)(struct x86_emulate_ctxt *ctxt);
  89         void (*exiting_smm)(struct x86_emulate_ctxt *ctxt);
  90 @@ -518,7 +519,8 @@ struct kvm_smram_state_32 {
  91         u32 reserved1[62];
  92         u32 smbase;
  93         u32 smm_revision;
  94 -       u32 reserved2[5];
  95 +       u32 reserved2[4];
  96 +       u32 int_shadow; /* KVM extension */
  97         u32 cr4; /* CR4 is not present in Intel/AMD SMRAM image */
  98         u32 reserved3[5];
  99
 100 @@ -566,6 +568,7 @@ static inline void __check_smram32_offsets(void)
 101         __CHECK_SMRAM32_OFFSET(smbase,          0xFEF8);
 102         __CHECK_SMRAM32_OFFSET(smm_revision,    0xFEFC);
 103         __CHECK_SMRAM32_OFFSET(reserved2,       0xFF00);
 104 +       __CHECK_SMRAM32_OFFSET(int_shadow,      0xFF10);
 105         __CHECK_SMRAM32_OFFSET(cr4,             0xFF14);
 106         __CHECK_SMRAM32_OFFSET(reserved3,       0xFF18);
 107         __CHECK_SMRAM32_OFFSET(ds,              0xFF2C);
 108 @@ -625,7 +628,7 @@ struct kvm_smram_state_64 {
 109         u64 io_restart_rsi;
 110         u64 io_restart_rdi;
 111         u32 io_restart_dword;
 112 -       u32 reserved1;
 113 +       u32 int_shadow;
 114         u8 io_inst_restart;
 115         u8 auto_hlt_restart;
 116         u8 reserved2[6];
 117 @@ -663,7 +666,6 @@ struct kvm_smram_state_64 {
 118         u64 gprs[16]; /* GPRS in a reversed "natural" X86 order (R15/R14/../RCX/RAX.) */
 119  };
 120
 121 -
 122  static inline void __check_smram64_offsets(void)
 123  {
 124  #define __CHECK_SMRAM64_OFFSET(field, offset) \
 125 @@ -684,7 +686,7 @@ static inline void __check_smram64_offsets(void)
 126         __CHECK_SMRAM64_OFFSET(io_restart_rsi,          0xFEB0);
 127         __CHECK_SMRAM64_OFFSET(io_restart_rdi,          0xFEB8);
 128         __CHECK_SMRAM64_OFFSET(io_restart_dword,        0xFEC0);
 129 -       __CHECK_SMRAM64_OFFSET(reserved1,               0xFEC4);
 130 +       __CHECK_SMRAM64_OFFSET(int_shadow,              0xFEC4);
 131         __CHECK_SMRAM64_OFFSET(io_inst_restart,         0xFEC8);
 132         __CHECK_SMRAM64_OFFSET(auto_hlt_restart,        0xFEC9);
 133         __CHECK_SMRAM64_OFFSET(reserved2,               0xFECA);
 134 diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
 135 index f5d3c0ff083d..e4573ac6ee94 100644
 136 --- a/arch/x86/kvm/x86.c
 137 +++ b/arch/x86/kvm/x86.c
 138 @@ -8192,6 +8192,11 @@ static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
 139         static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
 140  }
 141
 142 +static void emulator_set_int_shadow(struct x86_emulate_ctxt *ctxt, u8 shadow)
 143 +{
 144 +        static_call(kvm_x86_set_interrupt_shadow)(emul_to_vcpu(ctxt), shadow);
 145 +}
 146 +
 147  static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
 148  {
 149         return emul_to_vcpu(ctxt)->arch.hflags;
 150 @@ -8272,6 +8277,7 @@ static const struct x86_emulate_ops emulate_ops = {
 151         .guest_has_fxsr      = emulator_guest_has_fxsr,
 152         .guest_has_rdpid     = emulator_guest_has_rdpid,
 153         .set_nmi_mask        = emulator_set_nmi_mask,
 154 +       .set_int_shadow      = emulator_set_int_shadow,
 155         .get_hflags          = emulator_get_hflags,
 156         .exiting_smm         = emulator_exiting_smm,
 157         .leave_smm           = emulator_leave_smm,
 158 @@ -10191,6 +10197,8 @@ static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, struct kvm_smram_stat
 159         smram->cr4 = kvm_read_cr4(vcpu);
 160         smram->smm_revision = 0x00020000;
 161         smram->smbase = vcpu->arch.smbase;
 162 +
 163 +       smram->int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
 164  }
 165
 166  #ifdef CONFIG_X86_64
 167 @@ -10239,6 +10247,8 @@ static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, struct kvm_smram_stat
 168         enter_smm_save_seg_64(vcpu, &smram->ds, VCPU_SREG_DS);
 169         enter_smm_save_seg_64(vcpu, &smram->fs, VCPU_SREG_FS);
 170         enter_smm_save_seg_64(vcpu, &smram->gs, VCPU_SREG_GS);
 171 +
 172 +       smram->int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
 173  }
 174  #endif
 175
 176 @@ -10275,6 +10285,8 @@ static void enter_smm(struct kvm_vcpu *vcpu)
 177         kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
 178         kvm_rip_write(vcpu, 0x8000);
 179
 180 +       static_call(kvm_x86_set_interrupt_shadow)(vcpu, 0);
 181 +
 182         cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
 183         static_call(kvm_x86_set_cr0)(vcpu, cr0);
 184         vcpu->arch.cr0 = cr0;