From b10564cbaa8fe68ea5817f62b7c3c3e00c057eeb Mon Sep 17 00:00:00 2001
From: Alex Williamson <alex.williamson@hp.com>
Date: Thu, 16 Aug 2007 09:37:54 -0600
Subject: [PATCH] [IA64] Shrink vtlb size

Instrumenting thash_purge_all() shows a very low usage of vtlb
entries (21 at most).

This patch shrinks the default vtlb size from 512KB to 16KB
to optimize memory.  This also speeds up ptc_e emulation.
To improve the hash function, frequency of collision never changed
and there is no performance degradation.

Signed-off-by: Kouya Shimura <kouya@jp.fujitsu.com>
---
 xen/arch/ia64/vmx/vmmu.c    | 38 +++----------------------------------
 xen/arch/ia64/vmx/vtlb.c    | 17 ++++++++++++++---
 xen/include/asm-ia64/vmmu.h |  2 +-
 3 files changed, 18 insertions(+), 39 deletions(-)

diff --git a/xen/arch/ia64/vmx/vmmu.c b/xen/arch/ia64/vmx/vmmu.c
index b1eec86354..8ab640f542 100644
--- a/xen/arch/ia64/vmx/vmmu.c
+++ b/xen/arch/ia64/vmx/vmmu.c
@@ -32,9 +32,9 @@ static void __init parse_vtlb_size(char *s)
 
     if (sz > 0) {
         default_vtlb_sz = fls(sz - 1);
-        /* minimum 256KB (since calculated tag might be broken) */
-        if (default_vtlb_sz < 18)
-            default_vtlb_sz = 18;
+        /* minimum 16KB (for tag uniqueness) */
+        if (default_vtlb_sz < 14)
+            default_vtlb_sz = 14;
     }
 }
 
@@ -240,40 +240,8 @@ void machine_tlb_insert(struct vcpu *v, thash_data_t *tlb)
  */
 void machine_tlb_purge(u64 va, u64 ps)
 {
-//    u64       psr;
-//    psr = ia64_clear_ic();
     ia64_ptcl(va, ps << 2);
-//    ia64_set_psr(psr);
-//    ia64_srlz_i();
-//    return;
 }
-/*
-u64 machine_thash(u64 va)
-{
-    return ia64_thash(va);
-}
-
-u64 machine_ttag(u64 va)
-{
-    return ia64_ttag(va);
-}
-*/
-thash_data_t * vsa_thash(PTA vpta, u64 va, u64 vrr, u64 *tag)
-{
-    u64 index,pfn,rid,pfn_bits;
-    pfn_bits = vpta.size-5-8;
-    pfn = REGION_OFFSET(va)>>_REGION_PAGE_SIZE(vrr);
-    rid = _REGION_ID(vrr);
-    index = ((rid&0xff)<<pfn_bits)|(pfn&((1UL<<pfn_bits)-1));
-    *tag = ((rid>>8)&0xffff) | ((pfn >>pfn_bits)<<16);
-    return (thash_data_t *)((vpta.base<<PTA_BASE_SHIFT)+(index<<5));
-//    return ia64_call_vsa(PAL_VPS_THASH,va,vrr,vpta,0,0,0,0);
-}
-
-//u64 vsa_ttag(u64 va, u64 vrr)
-//{
-//    return ia64_call_vsa(PAL_VPS_TTAG,va,vrr,0,0,0,0,0);
-//}
 
 int vhpt_enabled(VCPU *vcpu, uint64_t vadr, vhpt_ref_t ref)
 {
diff --git a/xen/arch/ia64/vmx/vtlb.c b/xen/arch/ia64/vmx/vtlb.c
index 2a827ebcd6..7ac52dab89 100644
--- a/xen/arch/ia64/vmx/vtlb.c
+++ b/xen/arch/ia64/vmx/vtlb.c
@@ -286,6 +286,17 @@ u64 guest_vhpt_lookup(u64 iha, u64 *pte)
     return ret;
 }
 
+static thash_data_t * vtlb_thash(PTA vpta, u64 va, u64 vrr, u64 *tag)
+{
+    u64 index, pfn, rid;
+
+    pfn = REGION_OFFSET(va) >> _REGION_PAGE_SIZE(vrr);
+    rid = _REGION_ID(vrr);
+    index = (pfn ^ rid) & ((1UL << (vpta.size - 5)) - 1);
+    *tag = pfn ^ (rid << 39);
+    return (thash_data_t *)((vpta.base << PTA_BASE_SHIFT) + (index << 5));
+}
+
 /*
  *  purge software guest tlb
  */
@@ -308,7 +319,7 @@ static void vtlb_purge(VCPU *v, u64 va, u64 ps)
         size = PSIZE(rr_ps);
         vrr.ps = rr_ps;
         while (num) {
-            cur = vsa_thash(hcb->pta, curadr, vrr.rrval, &tag);
+            cur = vtlb_thash(hcb->pta, curadr, vrr.rrval, &tag);
             while (cur) {
                 if (cur->etag == tag && cur->ps == rr_ps)
                     cur->etag = 1UL << 63;
@@ -401,7 +412,7 @@ void vtlb_insert(VCPU *v, u64 pte, u64 itir, u64 va)
     vcpu_get_rr(v, va, &vrr.rrval);
     vrr.ps = itir_ps(itir);
     VMX(v, psbits[va >> 61]) |= (1UL << vrr.ps);
-    hash_table = vsa_thash(hcb->pta, va, vrr.rrval, &tag);
+    hash_table = vtlb_thash(hcb->pta, va, vrr.rrval, &tag);
     cch = hash_table;
     while (cch) {
         if (INVALID_TLB(cch)) {
@@ -639,7 +650,7 @@ thash_data_t *vtlb_lookup(VCPU *v, u64 va,int is_data)
         ps = __ffs(psbits);
         psbits &= ~(1UL << ps);
         vrr.ps = ps;
-        cch = vsa_thash(hcb->pta, va, vrr.rrval, &tag);
+        cch = vtlb_thash(hcb->pta, va, vrr.rrval, &tag);
         do {
             if (cch->etag == tag && cch->ps == ps)
                 return cch;
diff --git a/xen/include/asm-ia64/vmmu.h b/xen/include/asm-ia64/vmmu.h
index 3df4258b16..2b4ccfc6db 100644
--- a/xen/include/asm-ia64/vmmu.h
+++ b/xen/include/asm-ia64/vmmu.h
@@ -24,7 +24,7 @@
 #define XEN_TLBthash_H
 
 #define     MAX_CCN_DEPTH       (15)       // collision chain depth
-#define     DEFAULT_VTLB_SZ     (19) // 512K hash + 512K c-chain for VTLB
+#define     DEFAULT_VTLB_SZ     (14) // 16K hash + 16K c-chain for VTLB
 #define     DEFAULT_VHPT_SZ     (23) // 8M hash + 8M c-chain for VHPT
 #define     VTLB(v,_x)          (v->arch.vtlb._x)
 #define     VHPT(v,_x)          (v->arch.vhpt._x)
-- 
2.30.2