KVM: VMX: Split segments reload in vmx_load_host_state()

vmx_load_host_state() bundles fs, gs, ldt, and tss reloading into
one in the hope that it is infrequent. With smp guests, fs reloading is
frequent due to fs being used by threads.

Unbundle the reloads so reduce expensive gs reloads.
Signed-off-by: NLaurent Vivier <Laurent.Vivier@bull.net>
Signed-off-by: NAvi Kivity <avi@qumranet.com>

drivers/kvm/vmx.c

@@ -54,7 +54,8 @@ struct vcpu_vmx {
 	struct {
 		int           loaded;
 		u16           fs_sel, gs_sel, ldt_sel;
-		int           fs_gs_ldt_reload_needed;
+		int           gs_ldt_reload_needed;
+		int           fs_reload_needed;
 	}host_state;
 
 };
@@ -353,20 +354,21 @@ static void vmx_save_host_state(struct vcpu_vmx *vmx)
 	 * allow segment selectors with cpl > 0 or ti == 1.
 	 */
 	vmx->host_state.ldt_sel = read_ldt();
-	vmx->host_state.fs_gs_ldt_reload_needed = vmx->host_state.ldt_sel;
+	vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
 	vmx->host_state.fs_sel = read_fs();
-	if (!(vmx->host_state.fs_sel & 7))
+	if (!(vmx->host_state.fs_sel & 7)) {
 		vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
-	else {
+		vmx->host_state.fs_reload_needed = 0;
+	} else {
 		vmcs_write16(HOST_FS_SELECTOR, 0);
-		vmx->host_state.fs_gs_ldt_reload_needed = 1;
+		vmx->host_state.fs_reload_needed = 1;
 	}
 	vmx->host_state.gs_sel = read_gs();
 	if (!(vmx->host_state.gs_sel & 7))
 		vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
 	else {
 		vmcs_write16(HOST_GS_SELECTOR, 0);
-		vmx->host_state.fs_gs_ldt_reload_needed = 1;
+		vmx->host_state.gs_ldt_reload_needed = 1;
 	}
 
 #ifdef CONFIG_X86_64
@@ -396,9 +398,10 @@ static void vmx_load_host_state(struct vcpu_vmx *vmx)
 		return;
 
 	vmx->host_state.loaded = 0;
-	if (vmx->host_state.fs_gs_ldt_reload_needed) {
-		load_ldt(vmx->host_state.ldt_sel);
+	if (vmx->host_state.fs_reload_needed)
 		load_fs(vmx->host_state.fs_sel);
+	if (vmx->host_state.gs_ldt_reload_needed) {
+		load_ldt(vmx->host_state.ldt_sel);
 		/*
 		 * If we have to reload gs, we must take care to
 		 * preserve our gs base.
@@ -409,9 +412,8 @@ static void vmx_load_host_state(struct vcpu_vmx *vmx)
 		wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
 #endif
 		local_irq_restore(flags);
-
-		reload_tss();
 	}
+	reload_tss();
 	save_msrs(vmx->guest_msrs, vmx->save_nmsrs);
 	load_msrs(vmx->host_msrs, vmx->save_nmsrs);
 	if (msr_efer_need_save_restore(vmx))