Skip to contentSkip to conversation

spapr: Enable DD2.3 accelerated count cache flush in pseries-5.0 machine

PATCH In review David Gibson 4 participants 6 messages Jan 30, 2020 – Jan 31, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 spapr: Enable DD2.3 accelerated count cache flush in pseries-5.0 machine

For POWER9 DD2.2 cpus, the best current Spectre v2 indirect branch mitigation is "count cache disabled", which is configured with:
-machine cap-ibs=fixed-ccd
However, this option isn't available on DD2.3 CPUs with KVM, because they don't have the count cache disabled.

For POWER9 DD2.3 cpus, it is "count cache flush with assist", configured with:
-machine cap-ibs=workaround,cap-ccf-assist=on
However this option isn't available on DD2.2 CPUs with KVM, because they don't have the special CCF assist instruction this relies on.

On current machine types, we default to "count cache flush w/o assist", that is:
-machine cap-ibs=workaround,cap-ccf-assist=off
This runs, with mitigation on both DD2.2 and DD2.3 host cpus, but has a fairly significant performance impact.

It turns out we can do better. The special instruction that CCF assist uses to trigger a count cache flush is a no-op on earlier CPUs, rather than trapping or causing other badness. It doesn't, of itself, implement the mitigation, but *if* we have count-cache-disabled, then the count cache flush is unnecessary, and so using the count cache flush mitigation is harmless.

Therefore for the new pseries-5.0 machine type, enable cap-ccf-assist by default. Along with that, suppress throwing an error if cap-ccf-assist is selected but KVM doesn't support it, as long as KVM *is* giving us count-cache-disabled. To allow TCG to work out of the box, even though it doesn't implement the ccf flush assist, downgrade the error in that case to a warning. This matches several Spectre mitigations where we allow TCG to operate for debugging, since we don't really make guarantees about TCG security properties anyway.

While we're there, make the TCG warning for this case match that for other mitigations.

---
 hw/ppc/spapr.c      |  5 ++++-
 hw/ppc/spapr_caps.c | 26 ++++++++++++++++++++++----
 2 files changed, 26 insertions(+), 5 deletions(-)

I have put this into my ppc-for-5.0 tree already, and hope to send a pull request tomorrow (Jan 31).

hw/ppc/spapr.c +4 -1
index 02cf53fc5b..deaa44f1ab 100644
@@ -4397,7 +4397,7 @@ static void spapr_machine_class_init(ObjectClass *oc, void *data)
     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
-    smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
+    smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
     spapr_caps_add_properties(smc, &error_abort);
     smc->irq = &spapr_irq_dual;
     smc->dr_phb_enabled = true;
@@ -4465,8 +4465,11 @@ DEFINE_SPAPR_MACHINE(5_0, "5.0", true);
  */
 static void spapr_machine_4_2_class_options(MachineClass *mc)
 {
+    SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
+
     spapr_machine_5_0_class_options(mc);
     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
+    smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
 }
 
 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
hw/ppc/spapr_caps.c +22 -4
index 481dfd2a27..d0d4b32a40 100644
@@ -482,18 +482,36 @@ static void cap_large_decr_cpu_apply(SpaprMachineState *spapr,
 static void cap_ccf_assist_apply(SpaprMachineState *spapr, uint8_t val,
                                  Error **errp)
 {
+    Error *local_err = NULL;
     uint8_t kvm_val = kvmppc_get_cap_count_cache_flush_assist();
 
     if (tcg_enabled() && val) {
-        /* TODO - for now only allow broken for TCG */
-        error_setg(errp,
-"Requested count cache flush assist capability level not supported by tcg,"
-                   " try appending -machine cap-ccf-assist=off");
+        /* TCG doesn't implement anything here, but allow with a warning */
+        error_setg(&local_err,
+                   "TCG doesn't support requested feature, cap-ccf-assist=on");
     } else if (kvm_enabled() && (val > kvm_val)) {
+        uint8_t kvm_ibs = kvmppc_get_cap_safe_indirect_branch();
+
+        if (kvm_ibs == SPAPR_CAP_FIXED_CCD) {
+            /*
+             * If we don't have CCF assist on the host, the assist
+             * instruction is a harmless no-op.  It won't correctly
+             * implement the cache count flush *but* if we have
+             * count-cache-disabled in the host, that flush is
+             * unnnecessary.  So, specifically allow this case.  This
+             * allows us to have better performance on POWER9 DD2.3,
+             * while still working on POWER9 DD2.2 and POWER8 host
+             * cpus.
+             */
+            return;
+        }
         error_setg(errp,
 "Requested count cache flush assist capability level not supported by kvm,"
                    " try appending -machine cap-ccf-assist=off");
     }
+
+    if (local_err != NULL)
+        warn_report_err(local_err);
 }
 
 SpaprCapabilityInfo capability_table[SPAPR_CAP_NUM] = {
Signature · 1 line
2.24.1
#1 lore raw
Quoted 97 lines · David Gibson

For POWER9 DD2.2 cpus, the best current Spectre v2 indirect branch mitigation is "count cache disabled", which is configured with:
-machine cap-ibs=fixed-ccd
However, this option isn't available on DD2.3 CPUs with KVM, because they don't have the count cache disabled.

For POWER9 DD2.3 cpus, it is "count cache flush with assist", configured with:
-machine cap-ibs=workaround,cap-ccf-assist=on
However this option isn't available on DD2.2 CPUs with KVM, because they don't have the special CCF assist instruction this relies on.

On current machine types, we default to "count cache flush w/o assist", that is:
-machine cap-ibs=workaround,cap-ccf-assist=off
This runs, with mitigation on both DD2.2 and DD2.3 host cpus, but has a fairly significant performance impact.

It turns out we can do better. The special instruction that CCF assist uses to trigger a count cache flush is a no-op on earlier CPUs, rather than trapping or causing other badness. It doesn't, of itself, implement the mitigation, but *if* we have count-cache-disabled, then the count cache flush is unnecessary, and so using the count cache flush mitigation is harmless.

Therefore for the new pseries-5.0 machine type, enable cap-ccf-assist by default. Along with that, suppress throwing an error if cap-ccf-assist is selected but KVM doesn't support it, as long as KVM *is* giving us count-cache-disabled. To allow TCG to work out of the box, even though it doesn't implement the ccf flush assist, downgrade the error in that case to a warning. This matches several Spectre mitigations where we allow TCG to operate for debugging, since we don't really make guarantees about TCG security properties anyway.

While we're there, make the TCG warning for this case match that for other mitigations.

---
 hw/ppc/spapr.c      |  5 ++++-
 hw/ppc/spapr_caps.c | 26 ++++++++++++++++++++++----
 2 files changed, 26 insertions(+), 5 deletions(-)

I have put this into my ppc-for-5.0 tree already, and hope to send a pull request tomorrow (Jan 31).

hw/ppc/spapr.c +4 -1
index 02cf53fc5b..deaa44f1ab 100644
@@ -4397,7 +4397,7 @@ static void spapr_machine_class_init(ObjectClass *oc, void *data)
     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
-    smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
+    smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
     spapr_caps_add_properties(smc, &error_abort);
     smc->irq = &spapr_irq_dual;
     smc->dr_phb_enabled = true;
@@ -4465,8 +4465,11 @@ DEFINE_SPAPR_MACHINE(5_0, "5.0", true);
  */
 static void spapr_machine_4_2_class_options(MachineClass *mc)
 {
+    SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
+
     spapr_machine_5_0_class_options(mc);
     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
+    smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
 }
 
 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
hw/ppc/spapr_caps.c +10 -4
index 481dfd2a27..d0d4b32a40 100644
@@ -482,18 +482,36 @@ static void cap_large_decr_cpu_apply(SpaprMachineState *spapr,
 static void cap_ccf_assist_apply(SpaprMachineState *spapr, uint8_t val,
                                  Error **errp)
 {
+    Error *local_err = NULL;
     uint8_t kvm_val = kvmppc_get_cap_count_cache_flush_assist();
 
     if (tcg_enabled() && val) {
-        /* TODO - for now only allow broken for TCG */
-        error_setg(errp,
-"Requested count cache flush assist capability level not supported by tcg,"
-                   " try appending -machine cap-ccf-assist=off");
+        /* TCG doesn't implement anything here, but allow with a warning */
+        error_setg(&local_err,
+                   "TCG doesn't support requested feature, cap-ccf-assist=on");
     } else if (kvm_enabled() && (val > kvm_val)) {
+        uint8_t kvm_ibs = kvmppc_get_cap_safe_indirect_branch();
+
+        if (kvm_ibs == SPAPR_CAP_FIXED_CCD) {
+            /*
+             * If we don't have CCF assist on the host, the assist
+             * instruction is a harmless no-op.  It won't correctly

hw/ppc/spapr_caps.c static void cap_large_decr_cpu_apply(SpaprMachineState *spapr, · line 508

Quoted 10 lines
hw/ppc/spapr_caps.c +9 -0
+             * implement the cache count flush *but* if we have
+             * count-cache-disabled in the host, that flush is
+             * unnnecessary.  So, specifically allow this case.  This
+             * allows us to have better performance on POWER9 DD2.3,
+             * while still working on POWER9 DD2.2 and POWER8 host
+             * cpus.
+             */
+            return;
+        }
         error_setg(errp,

local_err? ...

hw/ppc/spapr_caps.c static void cap_large_decr_cpu_apply(SpaprMachineState *spapr, · line 514

Quoted 6 lines
hw/ppc/spapr_caps.c +3 -0
 "Requested count cache flush assist capability level not supported by kvm,"
                    " try appending -machine cap-ccf-assist=off");
     }
+
+    if (local_err != NULL)
+        warn_report_err(local_err);

... or why don't you put warn_report_err() in the first part of the "if" as this is the only place where it is used?

Thanks,
Laurent

#2 lore raw
Quoted 102 lines · On Thu, 30 Jan 2020 09:09:18 +0100 Laurent Vivier <[email protected]> wrote:
Quoted 101 lines · David Gibson

For POWER9 DD2.2 cpus, the best current Spectre v2 indirect branch mitigation is "count cache disabled", which is configured with:
-machine cap-ibs=fixed-ccd
However, this option isn't available on DD2.3 CPUs with KVM, because they don't have the count cache disabled.

For POWER9 DD2.3 cpus, it is "count cache flush with assist", configured with:
-machine cap-ibs=workaround,cap-ccf-assist=on
However this option isn't available on DD2.2 CPUs with KVM, because they don't have the special CCF assist instruction this relies on.

On current machine types, we default to "count cache flush w/o assist", that is:
-machine cap-ibs=workaround,cap-ccf-assist=off
This runs, with mitigation on both DD2.2 and DD2.3 host cpus, but has a fairly significant performance impact.

It turns out we can do better. The special instruction that CCF assist uses to trigger a count cache flush is a no-op on earlier CPUs, rather than trapping or causing other badness. It doesn't, of itself, implement the mitigation, but *if* we have count-cache-disabled, then the count cache flush is unnecessary, and so using the count cache flush mitigation is harmless.

Therefore for the new pseries-5.0 machine type, enable cap-ccf-assist by default. Along with that, suppress throwing an error if cap-ccf-assist is selected but KVM doesn't support it, as long as KVM *is* giving us count-cache-disabled. To allow TCG to work out of the box, even though it doesn't implement the ccf flush assist, downgrade the error in that case to a warning. This matches several Spectre mitigations where we allow TCG to operate for debugging, since we don't really make guarantees about TCG security properties anyway.

While we're there, make the TCG warning for this case match that for other mitigations.

---
 hw/ppc/spapr.c      |  5 ++++-
 hw/ppc/spapr_caps.c | 26 ++++++++++++++++++++++----
 2 files changed, 26 insertions(+), 5 deletions(-)

I have put this into my ppc-for-5.0 tree already, and hope to send a pull request tomorrow (Jan 31).

hw/ppc/spapr.c +4 -1
index 02cf53fc5b..deaa44f1ab 100644
@@ -4397,7 +4397,7 @@ static void spapr_machine_class_init(ObjectClass *oc, void *data)
     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
-    smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
+    smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
     spapr_caps_add_properties(smc, &error_abort);
     smc->irq = &spapr_irq_dual;
     smc->dr_phb_enabled = true;
@@ -4465,8 +4465,11 @@ DEFINE_SPAPR_MACHINE(5_0, "5.0", true);
  */
 static void spapr_machine_4_2_class_options(MachineClass *mc)
 {
+    SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
+
     spapr_machine_5_0_class_options(mc);
     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
+    smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
 }
 
 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
hw/ppc/spapr_caps.c +14 -4
index 481dfd2a27..d0d4b32a40 100644
@@ -482,18 +482,36 @@ static void cap_large_decr_cpu_apply(SpaprMachineState *spapr,
 static void cap_ccf_assist_apply(SpaprMachineState *spapr, uint8_t val,
                                  Error **errp)
 {
+    Error *local_err = NULL;
     uint8_t kvm_val = kvmppc_get_cap_count_cache_flush_assist();
 
     if (tcg_enabled() && val) {
-        /* TODO - for now only allow broken for TCG */
-        error_setg(errp,
-"Requested count cache flush assist capability level not supported by tcg,"
-                   " try appending -machine cap-ccf-assist=off");
+        /* TCG doesn't implement anything here, but allow with a warning */
+        error_setg(&local_err,
+                   "TCG doesn't support requested feature, cap-ccf-assist=on");
     } else if (kvm_enabled() && (val > kvm_val)) {
+        uint8_t kvm_ibs = kvmppc_get_cap_safe_indirect_branch();
+
+        if (kvm_ibs == SPAPR_CAP_FIXED_CCD) {
+            /*
+             * If we don't have CCF assist on the host, the assist
+             * instruction is a harmless no-op.  It won't correctly
+             * implement the cache count flush *but* if we have
+             * count-cache-disabled in the host, that flush is
+             * unnnecessary.  So, specifically allow this case.  This
+             * allows us to have better performance on POWER9 DD2.3,
Quoted 9 lines

hw/ppc/spapr_caps.c static void cap_large_decr_cpu_apply(SpaprMachineState *spapr, · line 508

Quoted 6 lines
hw/ppc/spapr_caps.c +5 -0
+             * while still working on POWER9 DD2.2 and POWER8 host
+             * cpus.
+             */
+            return;
+        }
         error_setg(errp,

local_err? ...

No. This is an error we do want to propagate to the caller, certainly not turning it into a warning.

Quoted 10 lines

hw/ppc/spapr_caps.c static void cap_large_decr_cpu_apply(SpaprMachineState *spapr, · line 514

Quoted 6 lines
hw/ppc/spapr_caps.c +3 -0
 "Requested count cache flush assist capability level not supported by kvm,"
                    " try appending -machine cap-ccf-assist=off");
     }
+
+    if (local_err != NULL)
+        warn_report_err(local_err);

... or why don't you put warn_report_err() in the first part of the "if" as this is the only place where it is used?

Yes and there you see that local_err isn't even needed in the first place. We should simply call warn_report().

Quoted 4 lines

Thanks,
Laurent

#3 lore raw
In reply to David Gibson
Quoted 80 lines · On Thu, 30 Jan 2020 12:26:22 +1100 David Gibson <[email protected]>…

For POWER9 DD2.2 cpus, the best current Spectre v2 indirect branch mitigation is "count cache disabled", which is configured with:
-machine cap-ibs=fixed-ccd
However, this option isn't available on DD2.3 CPUs with KVM, because they don't have the count cache disabled.

For POWER9 DD2.3 cpus, it is "count cache flush with assist", configured with:
-machine cap-ibs=workaround,cap-ccf-assist=on
However this option isn't available on DD2.2 CPUs with KVM, because they don't have the special CCF assist instruction this relies on.

On current machine types, we default to "count cache flush w/o assist", that is:
-machine cap-ibs=workaround,cap-ccf-assist=off
This runs, with mitigation on both DD2.2 and DD2.3 host cpus, but has a fairly significant performance impact.

It turns out we can do better. The special instruction that CCF assist uses to trigger a count cache flush is a no-op on earlier CPUs, rather than trapping or causing other badness. It doesn't, of itself, implement the mitigation, but *if* we have count-cache-disabled, then the count cache flush is unnecessary, and so using the count cache flush mitigation is harmless.

Therefore for the new pseries-5.0 machine type, enable cap-ccf-assist by default. Along with that, suppress throwing an error if cap-ccf-assist is selected but KVM doesn't support it, as long as KVM *is* giving us count-cache-disabled. To allow TCG to work out of the box, even though it doesn't implement the ccf flush assist, downgrade the error in that case to a warning. This matches several Spectre mitigations where we allow TCG to operate for debugging, since we don't really make guarantees about TCG security properties anyway.

While we're there, make the TCG warning for this case match that for other mitigations.

---
 hw/ppc/spapr.c      |  5 ++++-
 hw/ppc/spapr_caps.c | 26 ++++++++++++++++++++++----
 2 files changed, 26 insertions(+), 5 deletions(-)

I have put this into my ppc-for-5.0 tree already, and hope to send a pull request tomorrow (Jan 31).

hw/ppc/spapr.c +4 -1
index 02cf53fc5b..deaa44f1ab 100644
@@ -4397,7 +4397,7 @@ static void spapr_machine_class_init(ObjectClass *oc, void *data)
     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
-    smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
+    smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
     spapr_caps_add_properties(smc, &error_abort);
     smc->irq = &spapr_irq_dual;
     smc->dr_phb_enabled = true;
@@ -4465,8 +4465,11 @@ DEFINE_SPAPR_MACHINE(5_0, "5.0", true);
  */
 static void spapr_machine_4_2_class_options(MachineClass *mc)
 {
+    SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
+
     spapr_machine_5_0_class_options(mc);
     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
+    smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
 }
 
 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
hw/ppc/spapr_caps.c +1 -0
index 481dfd2a27..d0d4b32a40 100644
@@ -482,18 +482,36 @@ static void cap_large_decr_cpu_apply(SpaprMachineState *spapr,
 static void cap_ccf_assist_apply(SpaprMachineState *spapr, uint8_t val,
                                  Error **errp)
 {
+    Error *local_err = NULL;

hw/ppc/spapr_caps.c static void cap_large_decr_cpu_apply(SpaprMachineState *spapr, · line 491

Quoted 10 lines
hw/ppc/spapr_caps.c +3 -4
     uint8_t kvm_val = kvmppc_get_cap_count_cache_flush_assist();
 
     if (tcg_enabled() && val) {
-        /* TODO - for now only allow broken for TCG */
-        error_setg(errp,
-"Requested count cache flush assist capability level not supported by tcg,"
-                   " try appending -machine cap-ccf-assist=off");
+        /* TCG doesn't implement anything here, but allow with a warning */
+        error_setg(&local_err,
+                   "TCG doesn't support requested feature, cap-ccf-assist=on");

The only user for local_err is warn_report_err() below. It would be simpler to simply call warn_report() here.

Quoted 26 lines
hw/ppc/spapr_caps.c +18 -0
     } else if (kvm_enabled() && (val > kvm_val)) {
+        uint8_t kvm_ibs = kvmppc_get_cap_safe_indirect_branch();
+
+        if (kvm_ibs == SPAPR_CAP_FIXED_CCD) {
+            /*
+             * If we don't have CCF assist on the host, the assist
+             * instruction is a harmless no-op.  It won't correctly
+             * implement the cache count flush *but* if we have
+             * count-cache-disabled in the host, that flush is
+             * unnnecessary.  So, specifically allow this case.  This
+             * allows us to have better performance on POWER9 DD2.3,
+             * while still working on POWER9 DD2.2 and POWER8 host
+             * cpus.
+             */
+            return;
+        }
         error_setg(errp,
 "Requested count cache flush assist capability level not supported by kvm,"
                    " try appending -machine cap-ccf-assist=off");
     }
+
+    if (local_err != NULL)
+        warn_report_err(local_err);
 }
 
 SpaprCapabilityInfo capability_table[SPAPR_CAP_NUM] = {
#4 lore raw
Quoted 85 lines · Greg Kurz
Quoted 82 lines · On Thu, 30 Jan 2020 12:26:22 +1100 David Gibson <[email protected]>…

For POWER9 DD2.2 cpus, the best current Spectre v2 indirect branch mitigation is "count cache disabled", which is configured with:
-machine cap-ibs=fixed-ccd
However, this option isn't available on DD2.3 CPUs with KVM, because they don't have the count cache disabled.

For POWER9 DD2.3 cpus, it is "count cache flush with assist", configured with:
-machine cap-ibs=workaround,cap-ccf-assist=on
However this option isn't available on DD2.2 CPUs with KVM, because they don't have the special CCF assist instruction this relies on.

On current machine types, we default to "count cache flush w/o assist", that is:
-machine cap-ibs=workaround,cap-ccf-assist=off
This runs, with mitigation on both DD2.2 and DD2.3 host cpus, but has a fairly significant performance impact.

It turns out we can do better. The special instruction that CCF assist uses to trigger a count cache flush is a no-op on earlier CPUs, rather than trapping or causing other badness. It doesn't, of itself, implement the mitigation, but *if* we have count-cache-disabled, then the count cache flush is unnecessary, and so using the count cache flush mitigation is harmless.

Therefore for the new pseries-5.0 machine type, enable cap-ccf-assist by default. Along with that, suppress throwing an error if cap-ccf-assist is selected but KVM doesn't support it, as long as KVM *is* giving us count-cache-disabled. To allow TCG to work out of the box, even though it doesn't implement the ccf flush assist, downgrade the error in that case to a warning. This matches several Spectre mitigations where we allow TCG to operate for debugging, since we don't really make guarantees about TCG security properties anyway.

While we're there, make the TCG warning for this case match that for other mitigations.

---
 hw/ppc/spapr.c      |  5 ++++-
 hw/ppc/spapr_caps.c | 26 ++++++++++++++++++++++----
 2 files changed, 26 insertions(+), 5 deletions(-)

I have put this into my ppc-for-5.0 tree already, and hope to send a pull request tomorrow (Jan 31).

hw/ppc/spapr.c +4 -1
index 02cf53fc5b..deaa44f1ab 100644
@@ -4397,7 +4397,7 @@ static void spapr_machine_class_init(ObjectClass *oc, void *data)
     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
-    smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
+    smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
     spapr_caps_add_properties(smc, &error_abort);
     smc->irq = &spapr_irq_dual;
     smc->dr_phb_enabled = true;
@@ -4465,8 +4465,11 @@ DEFINE_SPAPR_MACHINE(5_0, "5.0", true);
  */
 static void spapr_machine_4_2_class_options(MachineClass *mc)
 {
+    SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
+
     spapr_machine_5_0_class_options(mc);
     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
+    smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
 }
 
 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
hw/ppc/spapr_caps.c +1 -0
index 481dfd2a27..d0d4b32a40 100644
@@ -482,18 +482,36 @@ static void cap_large_decr_cpu_apply(SpaprMachineState *spapr,
 static void cap_ccf_assist_apply(SpaprMachineState *spapr, uint8_t val,
                                  Error **errp)
 {
+    Error *local_err = NULL;
     uint8_t kvm_val = kvmppc_get_cap_count_cache_flush_assist();
 
Quoted 11 lines

hw/ppc/spapr_caps.c static void cap_large_decr_cpu_apply(SpaprMachineState *spapr, · line 491

Quoted 8 lines
hw/ppc/spapr_caps.c +3 -4
     if (tcg_enabled() && val) {
-        /* TODO - for now only allow broken for TCG */
-        error_setg(errp,
-"Requested count cache flush assist capability level not supported by tcg,"
-                   " try appending -machine cap-ccf-assist=off");
+        /* TCG doesn't implement anything here, but allow with a warning */
+        error_setg(&local_err,
+                   "TCG doesn't support requested feature, cap-ccf-assist=on");

The only user for local_err is warn_report_err() below. It would be simpler to simply call warn_report() here.

Yeah, fair enough. I was doing it this way for consistency with some of the other cases where we warn only on TCG, but there's not really any point. Revised in my tree.

Quoted 28 lines
Quoted 27 lines
hw/ppc/spapr_caps.c +18 -0
     } else if (kvm_enabled() && (val > kvm_val)) {
+        uint8_t kvm_ibs = kvmppc_get_cap_safe_indirect_branch();
+
+        if (kvm_ibs == SPAPR_CAP_FIXED_CCD) {
+            /*
+             * If we don't have CCF assist on the host, the assist
+             * instruction is a harmless no-op.  It won't correctly
+             * implement the cache count flush *but* if we have
+             * count-cache-disabled in the host, that flush is
+             * unnnecessary.  So, specifically allow this case.  This
+             * allows us to have better performance on POWER9 DD2.3,
+             * while still working on POWER9 DD2.2 and POWER8 host
+             * cpus.
+             */
+            return;
+        }
         error_setg(errp,
 "Requested count cache flush assist capability level not supported by kvm,"
                    " try appending -machine cap-ccf-assist=off");
     }
+
+    if (local_err != NULL)
+        warn_report_err(local_err);
 }
 
 SpaprCapabilityInfo capability_table[SPAPR_CAP_NUM] = {
Signature · 4 lines
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson
#5 lore raw
In reply to David Gibson
Quoted 37 lines · David Gibson

For POWER9 DD2.2 cpus, the best current Spectre v2 indirect branch mitigation is "count cache disabled", which is configured with:
-machine cap-ibs=fixed-ccd
However, this option isn't available on DD2.3 CPUs with KVM, because they don't have the count cache disabled.

For POWER9 DD2.3 cpus, it is "count cache flush with assist", configured with:
-machine cap-ibs=workaround,cap-ccf-assist=on
However this option isn't available on DD2.2 CPUs with KVM, because they don't have the special CCF assist instruction this relies on.

On current machine types, we default to "count cache flush w/o assist", that is:
-machine cap-ibs=workaround,cap-ccf-assist=off
This runs, with mitigation on both DD2.2 and DD2.3 host cpus, but has a fairly significant performance impact.

It turns out we can do better. The special instruction that CCF assist uses to trigger a count cache flush is a no-op on earlier CPUs, rather than trapping or causing other badness. It doesn't, of itself, implement the mitigation, but *if* we have count-cache-disabled, then the count cache flush is unnecessary, and so using the count cache flush mitigation is harmless.

Therefore for the new pseries-5.0 machine type, enable cap-ccf-assist by default. Along with that, suppress throwing an error if cap-ccf-assist is selected but KVM doesn't support it, as long as KVM *is* giving us count-cache-disabled. To allow TCG to work out of the box, even though it doesn't implement the ccf flush assist, downgrade the error in that case to a warning. This matches several Spectre mitigations where we allow TCG to operate for debugging, since we don't really make guarantees about TCG security properties anyway.

While we're there, make the TCG warning for this case match that for other mitigations.

Quoted 8 lines
---
 hw/ppc/spapr.c      |  5 ++++-
 hw/ppc/spapr_caps.c | 26 ++++++++++++++++++++++----
 2 files changed, 26 insertions(+), 5 deletions(-)

I have put this into my ppc-for-5.0 tree already, and hope to send a pull request tomorrow (Jan 31).

I tested this on a DD 2.2 machine, seems to work as designed.

I had to modify our spectre_v2 test slightly because it is confused about the sysfs reported mitigation not reflecting reality, but that is an existing issue.

cheers

# uname -r
5.5.0-gcc-8.2.0
# cat /proc/cpuinfo 
processor       : 0
cpu             : POWER9 (architected), altivec supported
clock           : 3000.000000MHz
revision        : 2.2 (pvr 004e 1202)

timebase : 512000000
platform : pSeries
model : IBM pSeries (emulated by qemu)
machine : CHRP IBM pSeries (emulated by qemu)
MMU : Radix

# grep . /sys/devices/system/cpu/vulnerabilities/*
/sys/devices/system/cpu/vulnerabilities/itlb_multihit:Not affected
/sys/devices/system/cpu/vulnerabilities/l1tf:Mitigation: RFI Flush, L1D private per thread
/sys/devices/system/cpu/vulnerabilities/mds:Not affected
/sys/devices/system/cpu/vulnerabilities/meltdown:Mitigation: RFI Flush, L1D private per thread
/sys/devices/system/cpu/vulnerabilities/spec_store_bypass:Mitigation: Kernel entry/exit barrier (eieio)
/sys/devices/system/cpu/vulnerabilities/tsx_async_abort:Not affected
/sys/devices/system/cpu/vulnerabilities/spectre_v1:Mitigation: __user pointer sanitization, ori31 speculation barrier enabled

/sys/devices/system/cpu/vulnerabilities/spectre_v2:Mitigation: Software count cache flush (hardware accelerated), Software link stack flush

# /spectre_v2 
test: spectre_v2
tags: git_version:v5.5-1-g0b2cecc02bf1
sysfs reports: 'Mitigation: Software count cache flush (hardware accelerated), Software link stack flush'
HW accelerated count cache flush on P9N DD2.2 => actually count cache disabled.
 PM_BR_PRED_CCACHE: result 2147483649 running/enabled 19406588726
PM_BR_MPRED_CCACHE: result 2147483649 running/enabled 19406580624
 PM_BR_PRED_PCACHE: result          0 running/enabled 19406574530
PM_BR_MPRED_PCACHE: result          0 running/enabled 19406571036
Miss percent 100 %
OK - Measured branch prediction rates match reported spectre v2 mitigation.
success: spectre_v2

# [ 47.183779][ C0] sysrq: Entering xmon
0:mon> di $_switch

c00000000040e280  7c0802a6      mflr    r0
c00000000040e284  f8010010      std     r0,16(r1)
c00000000040e288  f821fe21      stdu    r1,-480(r1)
c00000000040e28c  f9c100e0      std     r14,224(r1)
c00000000040e290  f9e100e8      std     r15,232(r1)
c00000000040e294  fa0100f0      std     r16,240(r1)
c00000000040e298  fa2100f8      std     r17,248(r1)
c00000000040e29c  fa410100      std     r18,256(r1)
c00000000040e2a0  fa610108      std     r19,264(r1)
c00000000040e2a4  fa810110      std     r20,272(r1)
c00000000040e2a8  faa10118      std     r21,280(r1)
c00000000040e2ac  fac10120      std     r22,288(r1)
c00000000040e2b0  fae10128      std     r23,296(r1)
c00000000040e2b4  fb010130      std     r24,304(r1)
c00000000040e2b8  fb210138      std     r25,312(r1)
c00000000040e2bc  fb410140      std     r26,320(r1)
c00000000040e2c0  fb610148      std     r27,328(r1)
c00000000040e2c4  fb810150      std     r28,336(r1)
c00000000040e2c8  fba10158      std     r29,344(r1)
c00000000040e2cc  fbc10160      std     r30,352(r1)
c00000000040e2d0  fbe10168      std     r31,360(r1)
c00000000040e2d4  f8010170      std     r0,368(r1)
c00000000040e2d8  7ee00026      mfcr    r23
c00000000040e2dc  fae101a0      std     r23,416(r1)
c00000000040e2e0  f8230000      std     r1,0(r3)
c00000000040e2e4  4bffdafd      bl      c00000000040bde0        # flush_count_cache+0x0/0x24a0

0:mon> di $flush_count_cache
c00000000040bde0 7d2802a6 mflr r9

# This is the link stack flush ...
c00000000040bde4 48000005 bl c00000000040bde8 # flush_count_cache+0x8/0x24a0
...
0:mon>
c00000000040be20 48000005 bl c00000000040be24 # flush_count_cache+0x44/0x24a0
...
0:mon>
c00000000040be60 48000005 bl c00000000040be64 # flush_count_cache+0x84/0x24a0
...
0:mon>
c00000000040bea0 48000005 bl c00000000040bea4 # flush_count_cache+0xc4/0x24a0
...
0:mon>

c00000000040bee0  48000005      bl      c00000000040bee4        # flush_count_cache+0x104/0x24a0
c00000000040bee4  4800001c      b       c00000000040bf00        # flush_count_cache+0x120/0x24a0
c00000000040bee8  60000000      nop
 ...
c00000000040bf00  7d2803a6      mtlr    r9
c00000000040bf04  60000000      nop
c00000000040bf08  39207fff      li      r9,32767
c00000000040bf0c  7d2903a6      mtctr   r9

# This is the HW accelerated count cache flush

c00000000040bf10  4c400420      bcctr-  2,lt
c00000000040bf14  4e800020      blr