- removing patch.fetchdecode-cache as it is included in the main branch, and does not apply cleanly anymore

This commit is contained in:
Christophe Bothamy 2002-06-27 14:39:03 +00:00
parent badef8cec8
commit c6f9ce353a

View File

@ -1,325 +0,0 @@
----------------------------------------------------------------------
Patch name: patche-fetchdecode-cache
Author: tld
Date: June, 3rd 2002
Detailed description:
3rd version of tld's fetchdecode cache
Patch was created with:
cvs diff -u
Apply patch to what version:
cvs checked out on June, 3rd 2002
Instructions:
To patch, go to main bochs directory.
Type "patch -p0 < THIS_PATCH_FILE".
----------------------------------------------------------------------
Index: config.h.in
===================================================================
RCS file: /cvsroot/bochs/bochs/config.h.in,v
retrieving revision 1.48
diff -u -r1.48 config.h.in
--- config.h.in 18 Apr 2002 01:00:53 -0000 1.48
+++ config.h.in 3 Jun 2002 20:31:02 -0000
@@ -546,6 +546,15 @@
#define BX_DYNAMIC_CPU_I386 0
#define BX_DYNAMIC_CPU_SPARC 0
+// caching of fetchdecode() calls
+#define BX_FETCHDECODE_CACHE 1
+
+#if BX_FETCHDECODE_CACHE
+ // The number of entries. MUST be a power of 2
+ #define BX_FDCACHE_SIZE 0x0800
+ #define BX_FDCACHE_MASK (BX_FDCACHE_SIZE-1)
+#endif // BX_FETCHDECODE_CACHE
+
#define BX_SUPPORT_FPU 0
#define BX_HAVE_GETENV 0
Index: cpu/cpu.cc
===================================================================
RCS file: /cvsroot/bochs/bochs/cpu/cpu.cc,v
retrieving revision 1.28
diff -u -r1.28 cpu.cc
--- cpu/cpu.cc 18 Apr 2002 00:22:19 -0000 1.28
+++ cpu/cpu.cc 3 Jun 2002 20:31:03 -0000
@@ -37,7 +37,12 @@
//unsigned counter[2] = { 0, 0 };
+#if BX_FETCHDECODE_CACHE
+ static unsigned long bx_fdcache_sel;
+ static unsigned long bx_fdcache_eip;
+ static Bit32u new_phy_addr;
+#endif // BX_FETCHDECODE_CACHE
#if BX_SIM_ID == 0 // only need to define once
// This array defines a look-up table for the even parity-ness
@@ -106,11 +111,16 @@
BX_CPU_C::cpu_loop(Bit32s max_instr_count)
{
unsigned ret;
- BxInstruction_t i;
+ BxInstruction_t *i;
unsigned maxisize;
Bit8u *fetch_ptr;
Boolean is_32;
+#if !BX_FETCHDECODE_CACHE
+ BxInstruction_t bxinstruction_dummy;
+ i = &bxinstruction_dummy;
+#endif // #if BX_FETCHDECODE_CACHE
+
#if BX_DEBUGGER
BX_CPU_THIS_PTR break_point = 0;
#ifdef MAGIC_BREAKPOINT
@@ -207,25 +217,53 @@
}
#endif
- is_32 = BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.d_b;
-
if (BX_CPU_THIS_PTR bytesleft == 0) {
prefetch();
}
fetch_ptr = BX_CPU_THIS_PTR fetch_ptr;
+#if BX_FETCHDECODE_CACHE
+ bx_fdcache_eip = new_phy_addr;
+ bx_fdcache_sel = bx_fdcache_eip & BX_FDCACHE_MASK;
+
+ i = &(BX_CPU_THIS_PTR fdcache_i[bx_fdcache_sel]);
+
+ if (BX_CPU_THIS_PTR fdcache_eip[bx_fdcache_sel] == bx_fdcache_eip) {
+ // HIT! ;^)
+ ret = 1; // success!
+ new_phy_addr += i->ilen;
+ } else {
+ // MISS :'(
+#endif // #if BX_FETCHDECODE_CACHE
+
maxisize = 16;
- if (BX_CPU_THIS_PTR bytesleft < 16)
+ if (BX_CPU_THIS_PTR bytesleft < 16) {
maxisize = BX_CPU_THIS_PTR bytesleft;
- ret = FetchDecode(fetch_ptr, &i, maxisize, is_32);
+ }
+ is_32 = BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.d_b;
+ ret = FetchDecode(fetch_ptr, i, maxisize, is_32);
+
+#if BX_FETCHDECODE_CACHE
+ // The instruction straddles a page boundary.
+ // Not storing such instructions in the cache is probably the
+ // easiest way to handle them
+ if (ret) {
+ BX_CPU_THIS_PTR fdcache_eip[bx_fdcache_sel] = bx_fdcache_eip;
+ new_phy_addr += i->ilen;
+ } else {
+ // Invalidate cache!
+ BX_CPU_THIS_PTR fdcache_eip[bx_fdcache_sel] = 0xFFFFFFFF;
+ }
+ }
+#endif // #if BX_FETCHDECODE_CACHE
if (ret) {
- if (i.ResolveModrm) {
+ if (i->ResolveModrm) {
// call method on BX_CPU_C object
- BX_CPU_CALL_METHOD(i.ResolveModrm, (&i));
+ BX_CPU_CALL_METHOD(i->ResolveModrm, (i));
}
- BX_CPU_THIS_PTR fetch_ptr += i.ilen;
- BX_CPU_THIS_PTR bytesleft -= i.ilen;
+ BX_CPU_THIS_PTR fetch_ptr += i->ilen;
+ BX_CPU_THIS_PTR bytesleft -= i->ilen;
fetch_decode_OK:
#if BX_DEBUGGER
@@ -239,34 +277,34 @@
}
#endif
- if (i.rep_used && (i.attr & BxRepeatable)) {
+ if (i->rep_used && (i->attr & BxRepeatable)) {
repeat_loop:
- if (i.attr & BxRepeatableZF) {
- if (i.as_32) {
+ if (i->attr & BxRepeatableZF) {
+ if (i->as_32) {
if (ECX != 0) {
- BX_CPU_CALL_METHOD(i.execute, (&i));
+ BX_CPU_CALL_METHOD(i->execute, (i));
ECX -= 1;
}
- if ((i.rep_used==0xf3) && (get_ZF()==0)) goto repeat_done;
- if ((i.rep_used==0xf2) && (get_ZF()!=0)) goto repeat_done;
+ if ((i->rep_used==0xf3) && (get_ZF()==0)) goto repeat_done;
+ if ((i->rep_used==0xf2) && (get_ZF()!=0)) goto repeat_done;
if (ECX == 0) goto repeat_done;
goto repeat_not_done;
}
else {
if (CX != 0) {
- BX_CPU_CALL_METHOD(i.execute, (&i));
+ BX_CPU_CALL_METHOD(i->execute, (i));
CX -= 1;
}
- if ((i.rep_used==0xf3) && (get_ZF()==0)) goto repeat_done;
- if ((i.rep_used==0xf2) && (get_ZF()!=0)) goto repeat_done;
+ if ((i->rep_used==0xf3) && (get_ZF()==0)) goto repeat_done;
+ if ((i->rep_used==0xf2) && (get_ZF()!=0)) goto repeat_done;
if (CX == 0) goto repeat_done;
goto repeat_not_done;
}
}
else { // normal repeat, no concern for ZF
- if (i.as_32) {
+ if (i->as_32) {
if (ECX != 0) {
- BX_CPU_CALL_METHOD(i.execute, (&i));
+ BX_CPU_CALL_METHOD(i->execute, (i));
ECX -= 1;
}
if (ECX == 0) goto repeat_done;
@@ -274,7 +312,7 @@
}
else { // 16bit addrsize
if (CX != 0) {
- BX_CPU_CALL_METHOD(i.execute, (&i));
+ BX_CPU_CALL_METHOD(i->execute, (i));
CX -= 1;
}
if (CX == 0) goto repeat_done;
@@ -302,12 +340,12 @@
repeat_done:
- BX_CPU_THIS_PTR eip += i.ilen;
+ BX_CPU_THIS_PTR eip += i->ilen;
}
else {
// non repeating instruction
- BX_CPU_THIS_PTR eip += i.ilen;
- BX_CPU_CALL_METHOD(i.execute, (&i));
+ BX_CPU_THIS_PTR eip += i->ilen;
+ BX_CPU_CALL_METHOD(i->execute, (i));
}
BX_CPU_THIS_PTR prev_eip = EIP; // commit new EIP
@@ -410,17 +448,22 @@
for (; j<16; j++) {
FetchBuffer[j] = *temp_ptr++;
}
- ret = FetchDecode(FetchBuffer, &i, 16, is_32);
+ ret = FetchDecode(FetchBuffer, i, 16, is_32);
if (ret==0)
BX_PANIC(("fetchdecode: cross boundary: ret==0"));
- if (i.ResolveModrm) {
- BX_CPU_CALL_METHOD(i.ResolveModrm, (&i));
+ if (i->ResolveModrm) {
+ BX_CPU_CALL_METHOD(i->ResolveModrm, (i));
}
- remain = i.ilen - remain;
+ remain = i->ilen - remain;
// note: eip has already been advanced to beginning of page
BX_CPU_THIS_PTR fetch_ptr = fetch_ptr + remain;
BX_CPU_THIS_PTR bytesleft -= remain;
+
+ #if BX_FETCHDECODE_CACHE
+ new_phy_addr += remain;
+ #endif // BX_FETCHDECODE_CACHE
+
//BX_CPU_THIS_PTR eip += remain;
BX_CPU_THIS_PTR eip = BX_CPU_THIS_PTR prev_eip;
goto fetch_decode_OK;
@@ -603,9 +646,12 @@
// cs:eIP
// prefetch QSIZE byte quantity aligned on corresponding boundary
Bit32u new_linear_addr;
- Bit32u new_phy_addr;
Bit32u temp_eip, temp_limit;
+#if !BX_FETCHDECODE_CACHE
+ Bit32u new_phy_addr;
+#endif // !BX_FETCHDECODE_CACHE
+
temp_eip = BX_CPU_THIS_PTR eip;
temp_limit = BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.limit_scaled;
@@ -664,7 +710,9 @@
BX_CPU_C::revalidate_prefetch_q(void)
{
Bit32u new_linear_addr, new_linear_page, new_linear_offset;
+#if !BX_FETCHDECODE_CACHE
Bit32u new_phy_addr;
+#endif // !BX_FETCHDECODE_CACHE
new_linear_addr = BX_CPU_THIS_PTR sregs[BX_SEG_REG_CS].cache.u.segment.base + BX_CPU_THIS_PTR eip;
Index: cpu/cpu.h
===================================================================
RCS file: /cvsroot/bochs/bochs/cpu/cpu.h,v
retrieving revision 1.19
diff -u -r1.19 cpu.h
--- cpu/cpu.h 18 Apr 2002 00:22:19 -0000 1.19
+++ cpu/cpu.h 3 Jun 2002 20:31:04 -0000
@@ -32,7 +32,6 @@
#include "cpu/lazy_flags.h"
-
#define BX_SREG_ES 0
#define BX_SREG_CS 1
#define BX_SREG_SS 2
@@ -1582,6 +1581,13 @@
bx_local_apic_c local_apic;
Boolean int_from_local_apic;
#endif
+
+ #if BX_FETCHDECODE_CACHE
+ Bit32u fdcache_eip[BX_FDCACHE_SIZE]; // will store operation's IP
+ // NOTE: This struct should really be aligned!
+ BxInstruction_t fdcache_i[BX_FDCACHE_SIZE]; // stores decoded instruction
+ #endif // #if BX_FETCHDECODE_CACHE
+
};
Index: memory/memory.cc
===================================================================
RCS file: /cvsroot/bochs/bochs/memory/memory.cc,v
retrieving revision 1.11
diff -u -r1.11 memory.cc
--- memory/memory.cc 3 Apr 2002 16:48:15 -0000 1.11
+++ memory/memory.cc 3 Jun 2002 20:31:05 -0000
@@ -49,6 +49,27 @@
a20addr = A20ADDR(addr);
BX_INSTR_PHY_WRITE(a20addr, len);
+#if BX_FETCHDECODE_CACHE
+ // NOTE: This piece should be put, if possible, where a write to the memory
+ // takes place.
+ // Here it trashes cache even for writes that would end up to ROM
+
+ // Invalidate instruction cache for written addresses
+ // Instructions can be up to 16 bytes long, so I have to trash up to 15 bytes
+ // before write address (costly!)
+ // I think it would NOT be safe to invalidate up to the last instruction
+ // before the write because there COULD be programs which use
+ // jump-in-the-middle-of-an-instruction schemes (esp. copyprotection
+ // schemes)
+ unsigned long bx_fdcache_idx = addr - 15;
+ for (int count = 15+len; count > 0; --count) {
+ if (cpu->fdcache_eip[bx_fdcache_idx & BX_FDCACHE_MASK] == bx_fdcache_idx) {
+ cpu->fdcache_eip[bx_fdcache_idx & BX_FDCACHE_MASK] = 0xFFFFFFFF;
+ }
+ ++bx_fdcache_idx;
+ }
+#endif // #if BX_FETCHDECODE_CACHE
+
#if BX_DEBUGGER
// (mch) Check for physical write break points, TODO
// (bbd) Each breakpoint should have an associated CPU#, TODO