[PATCH v2 0/2] x86/spec-ctrl: Minor fixes

[PATCH v2 0/2] x86/spec-ctrl: Minor fixes

[Andrew Cooper]

Patch 2 has been posted before, but ages ago and has been rebased over
subseqent XSAs.  Patch 1 is new.

Andrew Cooper (2):
  x86/spec-ctrl: Only adjust MSR_SPEC_CTRL for idle with legacy IBRS
  x86/spec-ctrl: Knobs for STIBP and PSFD, and follow hardware STIBP hint

 docs/misc/xen-command-line.pandoc      | 21 +++++++---
 xen/arch/x86/hvm/svm/vmcb.c            |  9 ++++
 xen/arch/x86/include/asm/cpufeatures.h |  2 +-
 xen/arch/x86/include/asm/spec_ctrl.h   |  5 ++-
 xen/arch/x86/spec_ctrl.c               | 75 +++++++++++++++++++++++++++++-----
 5 files changed, 93 insertions(+), 19 deletions(-)

-- 
2.11.0

[PATCH v2 1/2] x86/spec-ctrl: Only adjust MSR_SPEC_CTRL for idle with legacy IBRS

[Andrew Cooper]

Back at the time of the original Spectre-v2 fixes, it was recommended to clear
MSR_SPEC_CTRL when going idle.  This is because of the side effects on the
sibling thread caused by the microcode IBRS and STIBP implementations which
were retrofitted to existing CPUs.

However, there are no relevant cross-thread impacts for the hardware
IBRS/STIBP implementations, so this logic should not be used on Intel CPUs
supporting eIBRS, or any AMD CPUs; doing so only adds unnecessary latency to
the idle path.

Furthermore, there's no point playing with MSR_SPEC_CTRL in the idle paths if
SMT is disabled for other reasons.

Fixes: 8d03080d2a33 ("x86/spec-ctrl: Cease using thunk=lfence on AMD")
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
---
CC: Jan Beulich <JBeulich@suse.com>
CC: Roger Pau Monné <roger.pau@citrix.com>
CC: Wei Liu <wl@xen.org>

v2:
 * New
---
 xen/arch/x86/include/asm/cpufeatures.h |  2 +-
 xen/arch/x86/include/asm/spec_ctrl.h   |  5 +++--
 xen/arch/x86/spec_ctrl.c               | 10 ++++++++--
 3 files changed, 12 insertions(+), 5 deletions(-)

diff --git a/xen/arch/x86/include/asm/cpufeatures.h b/xen/arch/x86/include/asm/cpufeatures.h
index bd45a144ee78..493d338a085e 100644
--- a/xen/arch/x86/include/asm/cpufeatures.h
+++ b/xen/arch/x86/include/asm/cpufeatures.h
@@ -33,7 +33,7 @@ XEN_CPUFEATURE(SC_MSR_HVM,        X86_SYNTH(17)) /* MSR_SPEC_CTRL used by Xen fo
 XEN_CPUFEATURE(SC_RSB_PV,         X86_SYNTH(18)) /* RSB overwrite needed for PV */
 XEN_CPUFEATURE(SC_RSB_HVM,        X86_SYNTH(19)) /* RSB overwrite needed for HVM */
 XEN_CPUFEATURE(XEN_SELFSNOOP,     X86_SYNTH(20)) /* SELFSNOOP gets used by Xen itself */
-XEN_CPUFEATURE(SC_MSR_IDLE,       X86_SYNTH(21)) /* (SC_MSR_PV || SC_MSR_HVM) && default_xen_spec_ctrl */
+XEN_CPUFEATURE(SC_MSR_IDLE,       X86_SYNTH(21)) /* Clear MSR_SPEC_CTRL on idle */
 XEN_CPUFEATURE(XEN_LBR,           X86_SYNTH(22)) /* Xen uses MSR_DEBUGCTL.LBR */
 /* Bits 23,24 unused. */
 XEN_CPUFEATURE(SC_VERW_IDLE,      X86_SYNTH(25)) /* VERW used by Xen for idle */
diff --git a/xen/arch/x86/include/asm/spec_ctrl.h b/xen/arch/x86/include/asm/spec_ctrl.h
index 751355f471f4..7e83e0179fb9 100644
--- a/xen/arch/x86/include/asm/spec_ctrl.h
+++ b/xen/arch/x86/include/asm/spec_ctrl.h
@@ -78,7 +78,8 @@ static always_inline void spec_ctrl_enter_idle(struct cpu_info *info)
     uint32_t val = 0;
 
     /*
-     * Branch Target Injection:
+     * It is recommended in some cases to clear MSR_SPEC_CTRL when going idle,
+     * to avoid impacting sibling threads.
      *
      * Latch the new shadow value, then enable shadowing, then update the MSR.
      * There are no SMP issues here; only local processor ordering concerns.
@@ -114,7 +115,7 @@ static always_inline void spec_ctrl_exit_idle(struct cpu_info *info)
     uint32_t val = info->xen_spec_ctrl;
 
     /*
-     * Branch Target Injection:
+     * Restore MSR_SPEC_CTRL on exit from idle.
      *
      * Disable shadowing before updating the MSR.  There are no SMP issues
      * here; only local processor ordering concerns.
diff --git a/xen/arch/x86/spec_ctrl.c b/xen/arch/x86/spec_ctrl.c
index 1f275ad1fb5d..57f4fcb21398 100644
--- a/xen/arch/x86/spec_ctrl.c
+++ b/xen/arch/x86/spec_ctrl.c
@@ -1151,8 +1151,14 @@ void __init init_speculation_mitigations(void)
     /* (Re)init BSP state now that default_spec_ctrl_flags has been calculated. */
     init_shadow_spec_ctrl_state();
 
-    /* If Xen is using any MSR_SPEC_CTRL settings, adjust the idle path. */
-    if ( default_xen_spec_ctrl )
+    /*
+     * For microcoded IBRS only (i.e. Intel, pre eIBRS), it is recommended to
+     * clear MSR_SPEC_CTRL before going idle, to avoid impacting sibling
+     * threads.  Activate this if SMT is enabled, and Xen is using a non-zero
+     * MSR_SPEC_CTRL setting.
+     */
+    if ( boot_cpu_has(X86_FEATURE_IBRSB) && !(caps & ARCH_CAPS_IBRS_ALL) &&
+         hw_smt_enabled && default_xen_spec_ctrl )
         setup_force_cpu_cap(X86_FEATURE_SC_MSR_IDLE);
 
     xpti_init_default(caps);
-- 
2.11.0

[PATCH v2 2/2] x86/spec-ctrl: Knobs for STIBP and PSFD, and follow hardware STIBP hint

[Andrew Cooper]

STIBP and PSFD are slightly weird bits, because they're both implied by other
bits in MSR_SPEC_CTRL.  Add fine grain controls for them, and take the
implications into account when setting IBRS/SSBD.

Rearrange the IBPB text/variables/logic to keep all the MSR_SPEC_CTRL bits
together, for consistency.

However, AMD have a hardware hint CPUID bit recommending that STIBP be set
unilaterally.  This is advertised on Zen3, so follow the recommendation.
Furthermore, in such cases, set STIBP behind the guest's back for now.  This
has negligible overhead for the guest, but saves a WRMSR on vmentry.  This is
the only default change.

Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
---
CC: Jan Beulich <JBeulich@suse.com>
CC: Roger Pau Monné <roger.pau@citrix.com>
CC: Wei Liu <wl@xen.org>

v2:
 * Tweak comments/logic per suggestion.
 * Also set STIBP behind the guest's back to improve the vmentry path.
---
 docs/misc/xen-command-line.pandoc | 21 ++++++++++---
 xen/arch/x86/hvm/svm/vmcb.c       |  9 ++++++
 xen/arch/x86/spec_ctrl.c          | 65 +++++++++++++++++++++++++++++++++------
 3 files changed, 81 insertions(+), 14 deletions(-)

diff --git a/docs/misc/xen-command-line.pandoc b/docs/misc/xen-command-line.pandoc
index a92b7d228cae..da18172e50c5 100644
--- a/docs/misc/xen-command-line.pandoc
+++ b/docs/misc/xen-command-line.pandoc
@@ -2258,8 +2258,9 @@ By default SSBD will be mitigated at runtime (i.e `ssbd=runtime`).
 
 ### spec-ctrl (x86)
 > `= List of [ <bool>, xen=<bool>, {pv,hvm,msr-sc,rsb,md-clear}=<bool>,
->              bti-thunk=retpoline|lfence|jmp, {ibrs,ibpb,ssbd,eager-fpu,
->              l1d-flush,branch-harden,srb-lock,unpriv-mmio}=<bool> ]`
+>              bti-thunk=retpoline|lfence|jmp, {ibrs,ibpb,ssbd,psfd,
+>              eager-fpu,l1d-flush,branch-harden,srb-lock,
+>              unpriv-mmio}=<bool> ]`
 
 Controls for speculative execution sidechannel mitigations.  By default, Xen
 will pick the most appropriate mitigations based on compiled in support,
@@ -2309,9 +2310,10 @@ On hardware supporting IBRS (Indirect Branch Restricted Speculation), the
 If Xen is not using IBRS itself, functionality is still set up so IBRS can be
 virtualised for guests.
 
-On hardware supporting IBPB (Indirect Branch Prediction Barrier), the `ibpb=`
-option can be used to force (the default) or prevent Xen from issuing branch
-prediction barriers on vcpu context switches.
+On hardware supporting STIBP (Single Thread Indirect Branch Predictors), the
+`stibp=` option can be used to force or prevent Xen using the feature itself.
+By default, Xen will use STIBP when IBRS is in use (IBRS implies STIBP), and
+when hardware hints recommend using it as a blanket setting.
 
 On hardware supporting SSBD (Speculative Store Bypass Disable), the `ssbd=`
 option can be used to force or prevent Xen using the feature itself.  On AMD
@@ -2319,6 +2321,15 @@ hardware, this is a global option applied at boot, and not virtualised for
 guest use.  On Intel hardware, the feature is virtualised for guests,
 independently of Xen's choice of setting.
 
+On hardware supporting PSFD (Predictive Store Forwarding Disable), the `psfd=`
+option can be used to force or prevent Xen using the feature itself.  By
+default, Xen will not use PSFD.  PSFD is implied by SSBD, and SSBD is off by
+default.
+
+On hardware supporting IBPB (Indirect Branch Prediction Barrier), the `ibpb=`
+option can be used to force (the default) or prevent Xen from issuing branch
+prediction barriers on vcpu context switches.
+
 On all hardware, the `eager-fpu=` option can be used to force or prevent Xen
 from using fully eager FPU context switches.  This is currently implemented as
 a global control.  By default, Xen will choose to use fully eager context
diff --git a/xen/arch/x86/hvm/svm/vmcb.c b/xen/arch/x86/hvm/svm/vmcb.c
index 958309657799..0fc57dfd71cf 100644
--- a/xen/arch/x86/hvm/svm/vmcb.c
+++ b/xen/arch/x86/hvm/svm/vmcb.c
@@ -29,6 +29,7 @@
 #include <asm/hvm/support.h>
 #include <asm/hvm/svm/svm.h>
 #include <asm/hvm/svm/svmdebug.h>
+#include <asm/spec_ctrl.h>
 
 struct vmcb_struct *alloc_vmcb(void)
 {
@@ -176,6 +177,14 @@ static int construct_vmcb(struct vcpu *v)
             vmcb->_pause_filter_thresh = SVM_PAUSETHRESH_INIT;
     }
 
+    /*
+     * When default_xen_spec_ctrl simply SPEC_CTRL_STIBP, default this behind
+     * the back of the VM too.  Our SMT topology isn't accurate, the overhead
+     * is neglegable, and doing this saves a WRMSR on the vmentry path.
+     */
+    if ( default_xen_spec_ctrl == SPEC_CTRL_STIBP )
+        v->arch.msrs->spec_ctrl.raw = SPEC_CTRL_STIBP;
+
     return 0;
 }
 
diff --git a/xen/arch/x86/spec_ctrl.c b/xen/arch/x86/spec_ctrl.c
index 57f4fcb21398..efed24933d91 100644
--- a/xen/arch/x86/spec_ctrl.c
+++ b/xen/arch/x86/spec_ctrl.c
@@ -48,9 +48,13 @@ static enum ind_thunk {
     THUNK_LFENCE,
     THUNK_JMP,
 } opt_thunk __initdata = THUNK_DEFAULT;
+
 static int8_t __initdata opt_ibrs = -1;
+int8_t __initdata opt_stibp = -1;
+bool __read_mostly opt_ssbd;
+int8_t __initdata opt_psfd = -1;
+
 bool __read_mostly opt_ibpb = true;
-bool __read_mostly opt_ssbd = false;
 int8_t __read_mostly opt_eager_fpu = -1;
 int8_t __read_mostly opt_l1d_flush = -1;
 static bool __initdata opt_branch_harden = true;
@@ -172,12 +176,20 @@ static int __init cf_check parse_spec_ctrl(const char *s)
             else
                 rc = -EINVAL;
         }
+
+        /* Bits in MSR_SPEC_CTRL. */
         else if ( (val = parse_boolean("ibrs", s, ss)) >= 0 )
             opt_ibrs = val;
-        else if ( (val = parse_boolean("ibpb", s, ss)) >= 0 )
-            opt_ibpb = val;
+        else if ( (val = parse_boolean("stibp", s, ss)) >= 0 )
+            opt_stibp = val;
         else if ( (val = parse_boolean("ssbd", s, ss)) >= 0 )
             opt_ssbd = val;
+        else if ( (val = parse_boolean("psfd", s, ss)) >= 0 )
+            opt_psfd = val;
+
+        /* Misc settings. */
+        else if ( (val = parse_boolean("ibpb", s, ss)) >= 0 )
+            opt_ibpb = val;
         else if ( (val = parse_boolean("eager-fpu", s, ss)) >= 0 )
             opt_eager_fpu = val;
         else if ( (val = parse_boolean("l1d-flush", s, ss)) >= 0 )
@@ -376,7 +388,7 @@ static void __init print_details(enum ind_thunk thunk, uint64_t caps)
                "\n");
 
     /* Settings for Xen's protection, irrespective of guests. */
-    printk("  Xen settings: BTI-Thunk %s, SPEC_CTRL: %s%s%s%s, Other:%s%s%s%s%s\n",
+    printk("  Xen settings: BTI-Thunk %s, SPEC_CTRL: %s%s%s%s%s, Other:%s%s%s%s%s\n",
            thunk == THUNK_NONE      ? "N/A" :
            thunk == THUNK_RETPOLINE ? "RETPOLINE" :
            thunk == THUNK_LFENCE    ? "LFENCE" :
@@ -390,6 +402,9 @@ static void __init print_details(enum ind_thunk thunk, uint64_t caps)
            (!boot_cpu_has(X86_FEATURE_SSBD) &&
             !boot_cpu_has(X86_FEATURE_AMD_SSBD))     ? "" :
            (default_xen_spec_ctrl & SPEC_CTRL_SSBD)  ? " SSBD+" : " SSBD-",
+           (!boot_cpu_has(X86_FEATURE_PSFD) &&
+            !boot_cpu_has(X86_FEATURE_INTEL_PSFD))   ? "" :
+           (default_xen_spec_ctrl & SPEC_CTRL_PSFD)  ? " PSFD+" : " PSFD-",
            !(caps & ARCH_CAPS_TSX_CTRL)              ? "" :
            (opt_tsx & 1)                             ? " TSX+" : " TSX-",
            !cpu_has_srbds_ctrl                       ? "" :
@@ -980,10 +995,7 @@ void __init init_speculation_mitigations(void)
         if ( !has_spec_ctrl )
             printk(XENLOG_WARNING "?!? CET active, but no MSR_SPEC_CTRL?\n");
         else if ( opt_ibrs == -1 )
-        {
             opt_ibrs = ibrs = true;
-            default_xen_spec_ctrl |= SPEC_CTRL_IBRS | SPEC_CTRL_STIBP;
-        }
 
         if ( opt_thunk == THUNK_DEFAULT || opt_thunk == THUNK_RETPOLINE )
             thunk = THUNK_JMP;
@@ -1087,14 +1099,49 @@ void __init init_speculation_mitigations(void)
             setup_force_cpu_cap(X86_FEATURE_SC_MSR_HVM);
     }
 
-    /* If we have IBRS available, see whether we should use it. */
+    /* Figure out default_xen_spec_ctrl. */
     if ( has_spec_ctrl && ibrs )
+    {
+        /* IBRS implies STIBP.  */
+        if ( opt_stibp == -1 )
+            opt_stibp = 1;
+
         default_xen_spec_ctrl |= SPEC_CTRL_IBRS;
+    }
+
+    /*
+     * Use STIBP by default if the hardware hint is set.  Otherwise, leave it
+     * off as it a severe performance pentalty on pre-eIBRS Intel hardware
+     * where it was retrofitted in microcode.
+     */
+    if ( opt_stibp == -1 )
+        opt_stibp = !!boot_cpu_has(X86_FEATURE_STIBP_ALWAYS);
+
+    if ( opt_stibp && (boot_cpu_has(X86_FEATURE_STIBP) ||
+                       boot_cpu_has(X86_FEATURE_AMD_STIBP)) )
+        default_xen_spec_ctrl |= SPEC_CTRL_STIBP;
 
-    /* If we have SSBD available, see whether we should use it. */
     if ( opt_ssbd && (boot_cpu_has(X86_FEATURE_SSBD) ||
                       boot_cpu_has(X86_FEATURE_AMD_SSBD)) )
+    {
+        /* SSBD implies PSFD */
+        if ( opt_psfd == -1 )
+            opt_psfd = 1;
+
         default_xen_spec_ctrl |= SPEC_CTRL_SSBD;
+    }
+
+    /*
+     * Don't use PSFD by default.  AMD designed the predictor to
+     * auto-clear on privilege change.  PSFD is implied by SSBD, which is
+     * off by default.
+     */
+    if ( opt_psfd == -1 )
+        opt_psfd = 0;
+
+    if ( opt_psfd && (boot_cpu_has(X86_FEATURE_PSFD) ||
+                      boot_cpu_has(X86_FEATURE_INTEL_PSFD)) )
+        default_xen_spec_ctrl |= SPEC_CTRL_PSFD;
 
     /*
      * PV guests can create RSB entries for any linear address they control,
-- 
2.11.0

Re: [PATCH v2 1/2] x86/spec-ctrl: Only adjust MSR_SPEC_CTRL for idle with legacy IBRS

[Roger Pau Monné]

On Wed, Jun 29, 2022 at 07:45:07PM +0100, Andrew Cooper wrote:
> Back at the time of the original Spectre-v2 fixes, it was recommended to clear
> MSR_SPEC_CTRL when going idle.  This is because of the side effects on the
> sibling thread caused by the microcode IBRS and STIBP implementations which
> were retrofitted to existing CPUs.
> 
> However, there are no relevant cross-thread impacts for the hardware
> IBRS/STIBP implementations, so this logic should not be used on Intel CPUs
> supporting eIBRS, or any AMD CPUs; doing so only adds unnecessary latency to
> the idle path.
> 
> Furthermore, there's no point playing with MSR_SPEC_CTRL in the idle paths if
> SMT is disabled for other reasons.
> 
> Fixes: 8d03080d2a33 ("x86/spec-ctrl: Cease using thunk=lfence on AMD")
> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>

Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>

Thanks, Roger.

Re: [PATCH v2 2/2] x86/spec-ctrl: Knobs for STIBP and PSFD, and follow hardware STIBP hint

[Roger Pau Monné]

On Wed, Jun 29, 2022 at 07:45:08PM +0100, Andrew Cooper wrote:
> STIBP and PSFD are slightly weird bits, because they're both implied by other
> bits in MSR_SPEC_CTRL.  Add fine grain controls for them, and take the
> implications into account when setting IBRS/SSBD.
> 
> Rearrange the IBPB text/variables/logic to keep all the MSR_SPEC_CTRL bits
> together, for consistency.
> 
> However, AMD have a hardware hint CPUID bit recommending that STIBP be set
> unilaterally.  This is advertised on Zen3, so follow the recommendation.
> Furthermore, in such cases, set STIBP behind the guest's back for now.  This
> has negligible overhead for the guest, but saves a WRMSR on vmentry.  This is
> the only default change.
> 
> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
> Reviewed-by: Jan Beulich <jbeulich@suse.com>

Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>

Thanks, Roger.