[v2,6/6] x86/HVM: limit cache writeback overhead

Commit Message

Jan Beulich May 3, 2023, 9:47 a.m. UTC

There's no need to write back caches on all CPUs upon seeing a WBINVD
exit; ones that a vCPU hasn't run on since the last writeback (or since
it was started) can't hold data which may need writing back.

Signed-off-by: Jan Beulich <[e-mail]>
---
With us not running AMD IOMMUs in non-coherent ways, I wonder whether
svm_wbinvd_intercept() really needs to do anything (or whether it
couldn't check iommu_snoop just like VMX does, knowing that as of
c609108b2190 ["x86/shadow: make iommu_snoop usage consistent with
HAP's"] that's always set; this would largely serve as grep fodder then,
to make sure this code is updated once / when we do away with this
global variable, and it would be the penultimate step to being able to
fold SVM's and VT-x'es functions).
---
v2: Re-base over changes earlier in the series.

Patch

--- a/xen/arch/x86/hvm/hvm.c
+++ b/xen/arch/x86/hvm/hvm.c
@@ -537,6 +537,8 @@  void hvm_do_resume(struct vcpu *v)
         v->arch.hvm.inject_event.vector = HVM_EVENT_VECTOR_UNSET;
     }
 
+    __cpumask_set_cpu(v->processor, v->arch.hvm.cache_dirty_mask);
+
     if ( unlikely(v->arch.vm_event) && v->arch.monitor.next_interrupt_enabled )
     {
         struct x86_event info;
@@ -1592,6 +1594,10 @@  int hvm_vcpu_initialise(struct vcpu *v)
     if ( rc )
         goto fail6;
 
+    rc = -ENOMEM;
+    if ( !zalloc_cpumask_var(&v->arch.hvm.cache_dirty_mask) )
+        goto fail6;
+
     rc = ioreq_server_add_vcpu_all(d, v);
     if ( rc != 0 )
         goto fail6;
@@ -1621,6 +1627,7 @@  int hvm_vcpu_initialise(struct vcpu *v)
     hvm_vcpu_cacheattr_destroy(v);
  fail1:
     viridian_vcpu_deinit(v);
+    FREE_CPUMASK_VAR(v->arch.hvm.cache_dirty_mask);
     return rc;
 }
 
@@ -1628,6 +1635,8 @@  void hvm_vcpu_destroy(struct vcpu *v)
 {
     viridian_vcpu_deinit(v);
 
+    FREE_CPUMASK_VAR(v->arch.hvm.cache_dirty_mask);
+
     ioreq_server_remove_vcpu_all(v->domain, v);
 
     if ( hvm_altp2m_supported() )
--- a/xen/arch/x86/hvm/svm/svm.c
+++ b/xen/arch/x86/hvm/svm/svm.c
@@ -2363,8 +2363,14 @@  static void svm_vmexit_mce_intercept(
 
 static void cf_check svm_wbinvd_intercept(void)
 {
-    if ( cache_flush_permitted(current->domain) )
-        flush_all(FLUSH_CACHE_WRITEBACK);
+    struct vcpu *curr = current;
+
+    if ( !cache_flush_permitted(curr->domain) )
+        return;
+
+    flush_mask(curr->arch.hvm.cache_dirty_mask, FLUSH_CACHE_WRITEBACK);
+    cpumask_copy(curr->arch.hvm.cache_dirty_mask,
+                 cpumask_of(curr->processor));
 }
 
 static void svm_vmexit_do_invalidate_cache(struct cpu_user_regs *regs,
--- a/xen/arch/x86/hvm/vmx/vmx.c
+++ b/xen/arch/x86/hvm/vmx/vmx.c
@@ -3710,11 +3710,17 @@  static void vmx_do_extint(struct cpu_use
 
 static void cf_check vmx_wbinvd_intercept(void)
 {
-    if ( !cache_flush_permitted(current->domain) || iommu_snoop )
+    struct vcpu *curr = current;
+
+    if ( !cache_flush_permitted(curr->domain) || iommu_snoop )
         return;
 
     if ( cpu_has_wbinvd_exiting )
-        flush_all(FLUSH_CACHE_WRITEBACK);
+    {
+        flush_mask(curr->arch.hvm.cache_dirty_mask, FLUSH_CACHE_WRITEBACK);
+        cpumask_copy(curr->arch.hvm.cache_dirty_mask,
+                     cpumask_of(curr->processor));
+    }
     else
         wbnoinvd();
 }
--- a/xen/arch/x86/include/asm/hvm/vcpu.h
+++ b/xen/arch/x86/include/asm/hvm/vcpu.h
@@ -161,6 +161,8 @@  struct hvm_vcpu {
         struct svm_vcpu svm;
     };
 
+    cpumask_var_t       cache_dirty_mask;
+
     struct tasklet      assert_evtchn_irq_tasklet;
 
     struct nestedvcpu   nvcpu;