From 12a79192ee6a4ec5b69130ff3481ac64d93a9b25 Mon Sep 17 00:00:00 2001
From: mio <mio@lazym.io>
Date: Sat, 28 Jan 2023 22:04:43 +0100
Subject: [PATCH] Demand paging on Windows

---
 msvc/config-host.h             |  7 +++++++
 qemu/accel/tcg/translate-all.c | 18 ++++++++++++++++++
 qemu/include/hw/core/cpu.h     |  1 +
 qemu/include/tcg/tcg.h         |  5 +++++
 qemu/tcg/tcg.c                 | 18 ++++++++++++++++++
 5 files changed, 49 insertions(+)

diff --git a/msvc/config-host.h b/msvc/config-host.h
index 943d8593..bf750d2d 100644
--- a/msvc/config-host.h
+++ b/msvc/config-host.h
@@ -7,3 +7,10 @@
 #define CONFIG_CMPXCHG128 1
 // #define CONFIG_ATOMIC64 1
 #define CONFIG_PLUGIN 1
+
+// QEMU by default allocates (and commits) 1GB memory on Windows, and multiple Unicorn instances will result in OOM error easily.
+// Unfortunately, Windows doesn't have a similar demand paging feature like mmap(), therefore a workaround is to use tcg regions mechanism.
+// Note most Unicorn hacks (and even QEMU!) relies on the assumption that the translation memory won't run out and thus it might result
+// in some unexpected errors. If that is case, define to align with QEMU and Unicorn <= 2.0.1 behavior.
+// 
+// #define USE_STATIC_CODE_GEN_BUFFER
\ No newline at end of file
diff --git a/qemu/accel/tcg/translate-all.c b/qemu/accel/tcg/translate-all.c
index 91dd5ce7..aaebd4c5 100644
--- a/qemu/accel/tcg/translate-all.c
+++ b/qemu/accel/tcg/translate-all.c
@@ -869,6 +869,7 @@ static inline void *alloc_code_gen_buffer(struct uc_struct *uc)
     return buf;
 }
 #elif defined(_WIN32)
+#ifdef USE_STATIC_CODE_GEN_BUFFER
 static inline void *alloc_code_gen_buffer(struct uc_struct *uc)
 {
     TCGContext *tcg_ctx = uc->tcg_ctx;
@@ -876,6 +877,23 @@ static inline void *alloc_code_gen_buffer(struct uc_struct *uc)
     return VirtualAlloc(NULL, size, MEM_RESERVE | MEM_COMMIT,
                         PAGE_EXECUTE_READWRITE);
 }
+#else
+static inline void *alloc_code_gen_buffer(struct uc_struct *uc)
+{
+    TCGContext *tcg_ctx = uc->tcg_ctx;
+    size_t size = tcg_ctx->code_gen_buffer_size;
+
+    void* ptr =  VirtualAlloc(NULL, size, MEM_RESERVE,
+                        PAGE_EXECUTE_READWRITE);
+
+    // for prolog init
+    VirtualAlloc(ptr, 
+                 uc->qemu_real_host_page_size * UC_TCG_REGION_PAGES_COUNT,
+                 MEM_COMMIT, 
+                 PAGE_EXECUTE_READWRITE);
+    return ptr;
+}
+#endif
 void free_code_gen_buffer(struct uc_struct *uc)
 {
     TCGContext *tcg_ctx = uc->tcg_ctx;
diff --git a/qemu/include/hw/core/cpu.h b/qemu/include/hw/core/cpu.h
index d1b7c250..097a7f63 100644
--- a/qemu/include/hw/core/cpu.h
+++ b/qemu/include/hw/core/cpu.h
@@ -20,6 +20,7 @@
 #ifndef QEMU_CPU_H
 #define QEMU_CPU_H
 
+#include <stdint.h>
 #include "exec/hwaddr.h"
 #include "exec/memattrs.h"
 #include "qemu/bitmap.h"
diff --git a/qemu/include/tcg/tcg.h b/qemu/include/tcg/tcg.h
index f3643fe3..c7eb23ef 100644
--- a/qemu/include/tcg/tcg.h
+++ b/qemu/include/tcg/tcg.h
@@ -35,6 +35,11 @@
 #include "tcg-apple-jit.h"
 #include "qemu/int128.h"
 
+// Unicorn: Default region size for win32
+#if defined(_WIN32) && !defined(USE_STATIC_CODE_GEN_BUFFER)
+#define UC_TCG_REGION_PAGES_COUNT (128)
+#endif
+
 /* XXX: make safe guess about sizes */
 #define MAX_OP_PER_INSTR 266
 
diff --git a/qemu/tcg/tcg.c b/qemu/tcg/tcg.c
index 4910c67c..d5fba106 100644
--- a/qemu/tcg/tcg.c
+++ b/qemu/tcg/tcg.c
@@ -23,6 +23,8 @@
  */
 
 /* define it to use liveness analysis (better code) */
+#include "tcg/tcg.h"
+#include <stdio.h>
 #define USE_TCG_OPTIMIZATIONS
 
 #include "qemu/osdep.h"
@@ -406,6 +408,13 @@ static void tcg_region_assign(TCGContext *s, size_t curr_region)
     s->code_gen_buffer = start;
     s->code_gen_ptr = start;
     s->code_gen_buffer_size = (char *)end - (char *)start;
+#if defined(WIN32) && !defined(USE_STATIC_CODE_GEN_BUFFER)
+    VirtualAlloc(
+        s->code_gen_buffer, 
+        ROUND_UP(s->code_gen_buffer_size, s->uc->qemu_real_host_page_size), 
+        MEM_COMMIT, 
+        PAGE_EXECUTE_READWRITE);
+#endif
     memset(s->code_gen_buffer, 0x00, s->code_gen_buffer_size);
     s->code_gen_highwater = (char *)end - TCG_HIGHWATER;
 }
@@ -500,7 +509,11 @@ void tcg_region_init(TCGContext *tcg_ctx)
     size_t n_regions;
     size_t i;
 
+#if defined(WIN32) && !defined(USE_STATIC_CODE_GEN_BUFFER)
+    n_regions = size / (tcg_ctx->uc->qemu_real_host_page_size * UC_TCG_REGION_PAGES_COUNT);
+#else
     n_regions = 1;
+#endif
 
     /* The first region will be 'aligned - buf' bytes larger than the others */
     aligned = (void *)QEMU_ALIGN_PTR_UP(buf, page_size);
@@ -537,6 +550,11 @@ void tcg_region_init(TCGContext *tcg_ctx)
     }
 
     tcg_ctx->tree = g_tree_new(tb_tc_cmp);
+
+#if defined(WIN32) && !defined(USE_STATIC_CODE_GEN_BUFFER)
+    // Allocate a region immediately, or the highwater is not set correctly.
+    tcg_region_alloc(tcg_ctx);
+#endif
 }
 
 /*