A number of critical fixes painstakingly discovered in the slowest way possible.

4 files changed, 15 insertions, 9 deletions

diff --git a/lib/arm/arch-transform-dis.inc.h b/lib/arm/arch-transform-dis.inc.h
index 8569300..4c44913 100644
--- a/lib/arm/arch-transform-dis.inc.h
+++ b/lib/arm/arch-transform-dis.inc.h
@@ -12,6 +12,7 @@ static struct assemble_ctx tdctx_to_actx(const struct transform_dis_ctx *ctx) {
     }
     return (struct assemble_ctx) {
         ctx->rewritten_ptr_ptr,
+        (uint_tptr) (uintptr_t) ctx->rewritten_ptr_ptr,
         ctx->arch.pc_low_bit,
         cond
     };
@@ -121,7 +122,7 @@ void transform_dis_data(struct transform_dis_ctx *ctx, unsigned o0, unsigned o1,
 }
 
 static NOINLINE UNUSED
-void transform_dis_pcrel(struct transform_dis_ctx *ctx, uintptr_t dpc,
+void transform_dis_pcrel(struct transform_dis_ctx *ctx, uint_tptr dpc,
                          struct arch_pcrel_info info) {
 #ifdef TRANSFORM_DIS_VERBOSE
     printf("transform_dis_pcrel: (0x%llx) dpc=0x%llx reg=%x mode=%d\n",
@@ -149,7 +150,7 @@ void transform_dis_pcrel(struct transform_dis_ctx *ctx, uintptr_t dpc,
 }
 
 static NOINLINE UNUSED
-void transform_dis_branch(struct transform_dis_ctx *ctx, uintptr_t dpc, int cc) {
+void transform_dis_branch(struct transform_dis_ctx *ctx, uint_tptr dpc, int cc) {
 #ifdef TRANSFORM_DIS_VERBOSE
     printf("transform_dis (0x%llx): branch => 0x%llx\n",
            (unsigned long long) ctx->base.pc,
@@ -192,6 +193,7 @@ static void transform_dis_pre_dis(struct transform_dis_ctx *ctx) {
 static void transform_dis_post_dis(struct transform_dis_ctx *ctx) {
     if (ctx->arch.bccrel_p) {
         struct assemble_ctx actx = {&ctx->arch.bccrel_p,
+                                    (uint_tptr) (uintptr_t) ctx->arch.bccrel_p,
                                     /*thumb*/ true,
                                     ctx->arch.bccrel_bits};
         Bccrel(actx, *ctx->rewritten_ptr_ptr - ctx->arch.bccrel_p);
diff --git a/lib/arm/assemble.h b/lib/arm/assemble.h
index c0af020..2303a7a 100644
--- a/lib/arm/assemble.h
+++ b/lib/arm/assemble.h
@@ -3,6 +3,7 @@
 
 struct assemble_ctx {
     void **codep;
+    uint_tptr pc;
     bool thumb;
     int cond;
 };
@@ -100,6 +101,8 @@ static inline void Bccrel(struct assemble_ctx ctx, int offset) {
 }
 
 static inline void LDR_PC(struct assemble_ctx ctx, uint32_t dpc) {
+    if (ctx.pc & 2)
+        op16(ctx.codep, 0xbf00);
     if (ctx.thumb)
         op32(ctx.codep, 0xf000f8df);
     else
diff --git a/lib/arm/dis-thumb2.inc.h b/lib/arm/dis-thumb2.inc.h
index 0a4c12c..b8082f3 100644
--- a/lib/arm/dis-thumb2.inc.h
+++ b/lib/arm/dis-thumb2.inc.h
@@ -197,7 +197,7 @@ static INLINE void P(thumb2_do_it)(tdis_ctx ctx) {
 
 static INLINE void P(dis_thumb2)(tdis_ctx ctx) {
     ctx->base.op = unaligned_r32(ctx->base.ptr);
-    ctx->base.op_size = ctx->base.newop_size = 2;
+    ctx->base.op_size = ctx->base.newop_size = 4;
     /* LLVM likes to think about Thumb2 instructions the way the ARM manual
      * does - 15..0 15..0 rather than 31..0 as actually laid out in memory... */
     ctx->base.op = flip16(ctx->base.op);
diff --git a/lib/arm/jump-patch.h b/lib/arm/jump-patch.h
index b65a97d..8787070 100644
--- a/lib/arm/jump-patch.h
+++ b/lib/arm/jump-patch.h
@@ -2,17 +2,18 @@
 #include "dis.h"
 #include "arm/assemble.h"
 #define MAX_JUMP_PATCH_SIZE 8
+#define MAX_EXTENDED_PATCH_SIZE (MAX_JUMP_PATCH_SIZE+14)
 
-static inline int jump_patch_size(UNUSED uintptr_t pc,
-                                  UNUSED uintptr_t dpc,
+static inline int jump_patch_size(uint_tptr pc,
+                                  UNUSED uint_tptr dpc,
                                   UNUSED struct arch_dis_ctx arch,
                                   UNUSED bool force) {
-    return 8;
+    return (pc & 2) ? 10 : 8;
 }
 
-static inline void make_jump_patch(void **codep, UNUSED uintptr_t pc,
-                                   uintptr_t dpc,
+static inline void make_jump_patch(void **codep, uint_tptr pc,
+                                   uint_tptr dpc,
                                    struct arch_dis_ctx arch) {
-    struct assemble_ctx actx = {codep, arch.pc_low_bit, 0xe};
+    struct assemble_ctx actx = {codep, pc, arch.pc_low_bit, 0xe};
     LDR_PC(actx, dpc);
 }