drc: a bit more sophisticated f1 hack

[pcsx_rearmed.git] / libpcsxcore / new_dynarec / new_dynarec.c

diff --git a/libpcsxcore/new_dynarec/new_dynarec.c b/libpcsxcore/new_dynarec/new_dynarec.c
index 15b9383..0900736 100644 (file)
--- a/libpcsxcore/new_dynarec/new_dynarec.c
+++ b/libpcsxcore/new_dynarec/new_dynarec.c
@@ -29,10 +29,6 @@
 #ifdef _3DS
 #include <3ds_utils.h>
 #endif
-#ifdef VITA
-#include <psp2/kernel/sysmem.h>
-static int sceBlock;
-#endif
 
 #include "new_dynarec_config.h"
 #include "../psxhle.h"
@@ -228,7 +224,7 @@ static struct decoded_insn
   static void *copy;
   static int expirep;
   static u_int stop_after_jal;
-  static u_int f1_hack; // 0 - off, ~0 - capture address, else addr
+  static u_int f1_hack;
 
   int new_dynarec_hacks;
   int new_dynarec_hacks_pergame;
@@ -354,6 +350,14 @@ static void pass_args(int a0, int a1);
 static void emit_far_jump(const void *f);
 static void emit_far_call(const void *f);
 
+#ifdef VITA
+#include <psp2/kernel/sysmem.h>
+static int sceBlock;
+// note: this interacts with RetroArch's Vita bootstrap code: bootstrap/vita/sbrk.c
+extern int getVMBlock();
+int _newlib_vm_size_user = sizeof(*ndrc);
+#endif
+
 static void mprotect_w_x(void *start, void *end, int is_x)
 {
 #ifdef NO_WRITE_EXEC
@@ -6789,7 +6793,7 @@ static void new_dynarec_test(void)
     SysPrintf("linkage_arm* miscompilation/breakage detected.\n");
   }
 
-  SysPrintf("testing if we can run recompiled code...\n");
+  SysPrintf("testing if we can run recompiled code @%p...\n", out);
   ((volatile u_int *)out)[0]++; // make cache dirty
 
   for (i = 0; i < ARRAY_SIZE(ret); i++) {
@@ -6827,6 +6831,7 @@ void new_dynarec_clear_full(void)
   literalcount=0;
   stop_after_jal=0;
   inv_code_start=inv_code_end=~0;
+  hack_addr=0;
   f1_hack=0;
   // TLB
   for(n=0;n<4096;n++) ll_clear(jump_in+n);
@@ -6874,7 +6879,8 @@ void new_dynarec_init(void)
 #else
   #ifndef NO_WRITE_EXEC
   // not all systems allow execute in data segment by default
-  if (mprotect(ndrc, sizeof(ndrc->translation_cache) + sizeof(ndrc->tramp.ops),
+  // size must be 4K aligned for 3DS?
+  if (mprotect(ndrc, sizeof(*ndrc),
                PROT_READ | PROT_WRITE | PROT_EXEC) != 0)
     SysPrintf("mprotect() failed: %s\n", strerror(errno));
   #endif
@@ -6898,8 +6904,8 @@ void new_dynarec_cleanup(void)
   int n;
 #ifdef BASE_ADDR_DYNAMIC
   #ifdef VITA
-  sceKernelFreeMemBlock(sceBlock);
-  sceBlock = -1;
+  //sceKernelFreeMemBlock(sceBlock);
+  //sceBlock = -1;
   #else
   if (munmap(ndrc, sizeof(*ndrc)) < 0)
     SysPrintf("munmap() failed\n");
@@ -7041,11 +7047,11 @@ void new_dynarec_load_blocks(const void *save, int size)
   memcpy(&psxRegs.GPR, regs_save, sizeof(regs_save));
 }
 
-static void apply_hacks(void)
+static int apply_hacks(void)
 {
   int i;
   if (HACK_ENABLED(NDHACK_NO_COMPAT_HACKS))
-    return;
+    return 0;
   /* special hack(s) */
   for (i = 0; i < slen - 4; i++)
   {
@@ -7070,11 +7076,12 @@ static void apply_hacks(void)
     if (dops[i].itype == STORELR && dops[i].rs1 == 6
       && dops[i-1].itype == STORELR && dops[i-1].rs1 == 6)
     {
-      SysPrintf("F1 hack from %08x\n", start);
-      if (f1_hack == 0)
-        f1_hack = ~0u;
+      SysPrintf("F1 hack from %08x, old dst %08x\n", start, hack_addr);
+      f1_hack = 1;
+      return 1;
     }
   }
+  return 0;
 }
 
 int new_recompile_block(u_int addr)
@@ -7112,9 +7119,11 @@ int new_recompile_block(u_int addr)
     ll_add_flags(jump_in+page,start,state_rflags,(void *)beginning);
     return 0;
   }
-  else if (f1_hack == ~0u || (f1_hack != 0 && start == f1_hack)) {
+  else if (f1_hack && hack_addr == 0) {
     void *beginning = start_block();
     u_int page = get_page(start);
+    emit_movimm(start, 0);
+    emit_writeword(0, &hack_addr);
     emit_readword(&psxRegs.GPR.n.sp, 0);
     emit_readptr(&mem_rtab, 1);
     emit_shrimm(0, 12, 2);
@@ -7130,7 +7139,6 @@ int new_recompile_block(u_int addr)
 
     ll_add_flags(jump_in + page, start, state_rflags, beginning);
     SysPrintf("F1 hack to   %08x\n", start);
-    f1_hack = start;
     return 0;
   }
 
@@ -7162,9 +7170,9 @@ int new_recompile_block(u_int addr)
 
   /* Pass 1 disassembly */
 
-  for(i=0;!done;i++) {
-    dops[i].bt=0;
-    dops[i].ooo=0;
+  for (i = 0; !done; i++)
+  {
+    memset(&dops[i], 0, sizeof(dops[i]));
     op2=0;
     minimum_free_regs[i]=0;
     dops[i].opcode=op=source[i]>>26;
@@ -7604,12 +7612,12 @@ int new_recompile_block(u_int addr)
           do_in_intrp=1;
         }
       }
-      if(do_in_intrp) {
-        dops[i-1].rs1=CCREG;
-        dops[i-1].rs2=dops[i-1].rt1=dops[i-1].rt2=0;
-        ba[i-1]=-1;
-        dops[i-1].itype=INTCALL;
-        done=2;
+      if (do_in_intrp) {
+        memset(&dops[i-1], 0, sizeof(dops[i-1]));
+        dops[i-1].itype = INTCALL;
+        dops[i-1].rs1 = CCREG;
+        ba[i-1] = -1;
+        done = 2;
         i--; // don't compile the DS
       }
     }
@@ -7658,7 +7666,7 @@ int new_recompile_block(u_int addr)
   }
   assert(slen>0);
 
-  apply_hacks();
+  int clear_hack_addr = apply_hacks();
 
   /* Pass 2 - Register dependencies and branch targets */
 
@@ -7718,7 +7726,10 @@ int new_recompile_block(u_int addr)
         current.u=branch_unneeded_reg[i]&~((1LL<<dops[i+1].rs1)|(1LL<<dops[i+1].rs2));
         current.u&=~((1LL<<dops[i].rs1)|(1LL<<dops[i].rs2));
         current.u|=1;
-      } else { SysPrintf("oops, branch at end of block with no delay slot\n");abort(); }
+      } else {
+        SysPrintf("oops, branch at end of block with no delay slot @%08x\n", start + i*4);
+        abort();
+      }
     }
     dops[i].is_ds=ds;
     if(ds) {
@@ -9195,6 +9206,11 @@ int new_recompile_block(u_int addr)
       instr_addr[i] = out;
       assem_debug("<->\n");
       drc_dbg_emit_do_cmp(i, ccadj[i]);
+      if (clear_hack_addr) {
+        emit_movimm(0, 0);
+        emit_writeword(0, &hack_addr);
+        clear_hack_addr = 0;
+      }
 
       // load regs
       if(regs[i].regmap_entry[HOST_CCREG]==CCREG&&regs[i].regmap[HOST_CCREG]!=CCREG)