x86 vmx: Accelerate VLAPIC EOI writes

Our testing indicates that most apic accesses are eoi writes. This
patch accelerate guest EOI emulation utilizing HW VM Exit
information.

Without this patch, xentrace shows the apci access average tsc costs
is ~7.8k in our case and it down to ~3k with it. We also save 3% cpu
in our case.

From: Yang Zhang <yang.zhang@intel.com>
Signed-off-by: Keir Fraser <keir.fraser@citrix.com>

diff --git a/xen/arch/x86/hvm/vmx/vmx.c b/xen/arch/x86/hvm/vmx/vmx.c
index df8cccba525077d81abc9b84df6042b812513485..477f2e26b2257c3058bc90456dcd9738da51a21f 100644 (file)
--- a/xen/arch/x86/hvm/vmx/vmx.c
+++ b/xen/arch/x86/hvm/vmx/vmx.c
@@ -2270,6 +2270,26 @@ static void vmx_vmexit_ud_intercept(struct cpu_user_regs *regs)
     }
 }
 
+static int vmx_handle_eoi_write(void)
+{
+    unsigned long exit_qualification = __vmread(EXIT_QUALIFICATION);
+
+    /*
+     * 1. Must be a linear access data write.
+     * 2. Data write must be to the EOI register.
+     */
+    if ( (((exit_qualification >> 12) & 0xf) == 1) &&
+         ((exit_qualification & 0xfff) == APIC_EOI) )
+    {
+        int inst_len = __get_instruction_length(); /* Safe: APIC data write */
+        __update_guest_eip(inst_len);
+        vlapic_EOI_set(vcpu_vlapic(current));
+        return 1;
+    }
+
+    return 0;
+}
+
 asmlinkage void vmx_vmexit_handler(struct cpu_user_regs *regs)
 {
     unsigned int exit_reason, idtv_info;
@@ -2572,8 +2592,12 @@ asmlinkage void vmx_vmexit_handler(struct cpu_user_regs *regs)
     case EXIT_REASON_TPR_BELOW_THRESHOLD:
         break;
 
-    case EXIT_REASON_IO_INSTRUCTION:
     case EXIT_REASON_APIC_ACCESS:
+        if ( !vmx_handle_eoi_write() && !handle_mmio() )
+            vmx_inject_hw_exception(TRAP_gp_fault, 0);
+        break;
+
+    case EXIT_REASON_IO_INSTRUCTION:
         if ( !handle_mmio() )
             vmx_inject_hw_exception(TRAP_gp_fault, 0);
         break;