From 0e60d00e1986c0d6db070168e7576b64d5582ebf Mon Sep 17 00:00:00 2001
From: "awilliam@xenbuild2.aw" <awilliam@xenbuild2.aw>
Date: Thu, 25 Jan 2007 14:58:41 -0700
Subject: [PATCH] [IA64] Optimize vmx_vcpu_thash()

Implement in assembly

Signed-off-by: Zhang Xin <xing.z.zhang@intel.com>
---
 xen/arch/ia64/asm-offsets.c       |  1 +
 xen/arch/ia64/vmx/optvfault.S     | 60 +++++++++++++++++++++++++++++++
 xen/arch/ia64/vmx/vmx_ivt.S       |  2 ++
 xen/include/asm-ia64/vmx_mm_def.h |  4 ++-
 4 files changed, 66 insertions(+), 1 deletion(-)

diff --git a/xen/arch/ia64/asm-offsets.c b/xen/arch/ia64/asm-offsets.c
index c10cac45c7..6105406441 100644
--- a/xen/arch/ia64/asm-offsets.c
+++ b/xen/arch/ia64/asm-offsets.c
@@ -200,6 +200,7 @@ void foo(void)
 	DEFINE(IA64_VPD_BASE_OFFSET, offsetof (struct vcpu, arch.privregs));
 	DEFINE(IA64_VPD_VIFS_OFFSET, offsetof (mapped_regs_t, ifs));
  	DEFINE(IA64_VLSAPIC_INSVC_BASE_OFFSET, offsetof (struct vcpu, arch.insvc[0]));
+	DEFINE(IA64_VPD_VPTA_OFFSET, offsetof (struct mapped_regs, pta));
 	DEFINE(IA64_VPD_CR_VPTA_OFFSET, offsetof (cr_t, pta));
 	DEFINE(XXX_THASH_SIZE, sizeof (thash_data_t));
 
diff --git a/xen/arch/ia64/vmx/optvfault.S b/xen/arch/ia64/vmx/optvfault.S
index 6163437e5b..cf827ed98b 100644
--- a/xen/arch/ia64/vmx/optvfault.S
+++ b/xen/arch/ia64/vmx/optvfault.S
@@ -15,6 +15,7 @@
 #include <asm/vmx_vpd.h>
 #include <asm/vmx_pal_vsa.h>
 #include <asm/asm-offsets.h>
+#include <asm-ia64/vmx_mm_def.h>
 
 #define ACCE_MOV_FROM_AR
 #define ACCE_MOV_FROM_RR
@@ -22,6 +23,7 @@
 #define ACCE_RSM
 #define ACCE_SSM
 #define ACCE_MOV_TO_PSR
+#define ACCE_THASH
 
 //mov r1=ar3
 GLOBAL_ENTRY(vmx_asm_mov_from_ar)
@@ -418,6 +420,64 @@ ENTRY(vmx_asm_dispatch_vexirq)
     br.many vmx_dispatch_vexirq
 END(vmx_asm_dispatch_vexirq)
 
+// thash
+// TODO: add support when pta.vf = 1
+GLOBAL_ENTRY(vmx_asm_thash)
+#ifndef ACCE_THASH
+    br.many vmx_virtualization_fault_back
+#endif
+    extr.u r17=r25,20,7		// get r3 from opcode in r25 
+    extr.u r18=r25,6,7		// get r1 from opcode in r25
+    movl r20=asm_mov_from_reg
+    ;;
+    adds r30=vmx_asm_thash_back1-asm_mov_from_reg,r20
+    shladd r17=r17,4,r20	// get addr of MOVE_FROM_REG(r17)
+    adds r16=IA64_VPD_BASE_OFFSET,r21	// get vcpu.arch.priveregs
+    ;;
+    mov r24=b0
+    ;;
+    ld8 r16=[r16]		// get VPD addr
+    mov b0=r17
+    br.many b0			// r19 return value
+    ;;                                                     
+vmx_asm_thash_back1:
+    shr.u r23=r19,61		// get RR number
+    adds r25=VCPU_VRR0_OFS,r21	// get vcpu->arch.arch_vmx.vrr[0]'s addr
+    adds r16=IA64_VPD_VPTA_OFFSET,r16	// get vpta 
+    ;;
+    shladd r27=r23,3,r25	// get vcpu->arch.arch_vmx.vrr[r23]'s addr
+    ld8 r17=[r16]		// get PTA
+    mov r26=1
+    ;;
+    extr.u r29=r17,2,6		// get pta.size
+    ld8 r25=[r27]		// get vcpu->arch.arch_vmx.vrr[r23]'s value
+    ;;
+    extr.u r25=r25,2,6		// get rr.ps
+    shl r22=r26,r29		// 1UL << pta.size
+    ;;
+    shr.u r23=r19,r25		// vaddr >> rr.ps
+    adds r26=3,r29		// pta.size + 3 
+    shl r27=r17,3		// pta << 3 
+    ;;
+    shl r23=r23,3		// (vaddr >> rr.ps) << 3
+    shr.u r27=r27,r26		// (pta << 3) >> (pta.size+3)
+    movl r16=VRN_MASK
+    ;;
+    adds r22=-1,r22		// (1UL << pta.size) - 1
+    shl r27=r27,r29		// ((pta<<3)>>(pta.size+3))<<pta.size
+    and r19=r19,r16		// vaddr & VRN_MASK
+    ;;
+    and r22=r22,r23		// vhpt_offset 
+    or r19=r19,r27		// (vadr&VRN_MASK) |(((pta<<3)>>(pta.size + 3))<<pta.size) 
+    adds r26=asm_mov_to_reg-asm_mov_from_reg,r20
+    ;;
+    or r19=r19,r22		// calc pval
+    shladd r17=r18,4,r26
+    adds r30=vmx_resume_to_guest-asm_mov_from_reg,r20
+    ;;
+    mov b0=r17
+    br.many b0
+END(vmx_asm_thash)
 
 #define MOV_TO_REG0	\
 {;			\
diff --git a/xen/arch/ia64/vmx/vmx_ivt.S b/xen/arch/ia64/vmx/vmx_ivt.S
index 755539c09c..cf9b554298 100644
--- a/xen/arch/ia64/vmx/vmx_ivt.S
+++ b/xen/arch/ia64/vmx/vmx_ivt.S
@@ -795,12 +795,14 @@ ENTRY(vmx_virtualization_fault)
     cmp.eq p9,p0=EVENT_RSM,r24
     cmp.eq p10,p0=EVENT_SSM,r24
     cmp.eq p11,p0=EVENT_MOV_TO_PSR,r24
+    cmp.eq p12,p0=EVENT_THASH,r24 
     (p6) br.dptk.many vmx_asm_mov_from_ar
     (p7) br.dptk.many vmx_asm_mov_from_rr
     (p8) br.dptk.many vmx_asm_mov_to_rr
     (p9) br.dptk.many vmx_asm_rsm
     (p10) br.dptk.many vmx_asm_ssm
     (p11) br.dptk.many vmx_asm_mov_to_psr
+    (p12) br.dptk.many vmx_asm_thash
     ;;
 vmx_virtualization_fault_back:
     mov r19=37
diff --git a/xen/include/asm-ia64/vmx_mm_def.h b/xen/include/asm-ia64/vmx_mm_def.h
index 21cf118b54..4736908fda 100644
--- a/xen/include/asm-ia64/vmx_mm_def.h
+++ b/xen/include/asm-ia64/vmx_mm_def.h
@@ -103,7 +103,7 @@
 #define VA_MATTR_WC     0x6
 #define VA_MATTR_NATPAGE    0x7
 
-#define VRN_MASK        0xe000000000000000L
+#define VRN_MASK        0xe000000000000000
 #define PTA_BASE_MASK       0x3fffffffffffL
 #define PTA_BASE_SHIFT      15
 #define VHPT_OFFSET_MASK    0x7fff
@@ -114,6 +114,7 @@
 #define HPA_MAPPING_ATTRIBUTE   0x61  //ED:0;AR:0;PL:0;D:1;A:1;P:1
 #define VPN_2_VRN(vpn)  ((vpn << PPN_SHIFT) >> IA64_VRN_SHIFT)
 
+#ifndef __ASSEMBLY__
 typedef enum { INSTRUCTION, DATA, REGISTER } miss_type;
 
 //typedef enum { MVHPT, STLB } vtlb_loc_type_t;
@@ -169,5 +170,6 @@ bits_v(uint64_t v, uint32_t bs, uint32_t be)
                   "M" ((len)));                                 \
          ret;                                                   \
  })
+#endif
 
 #endif
-- 
2.30.2