Fix finish command for PNaCl.

gdb/frame.c

Index: gdb/frame.c
diff --git a/gdb/frame.c b/gdb/frame.c
index b7698733c2aea38d1a2b0c1329325980c5008521..2df99c7d0a136b365433a0667c56064e10cf2328 100644
--- a/gdb/frame.c
+++ b/gdb/frame.c
@@ -448,6 +448,7 @@ int
 frame_id_eq (struct frame_id l, struct frame_id r)
 {
   int eq;
+  uint32_t max_addr = (uint32_t) -1;

[eaeltsin, 2013/08/06 18:14:42]

const or #define?

if you use this for &, then either provide a synonym like 'addr_mask', or use %
instead.

[halyavin, 2013/08/07 08:04:25]

I used conversion instead of &.

 
   if (!l.stack_addr_p && l.special_addr_p
       && !r.stack_addr_p && r.special_addr_p)
@@ -461,7 +462,9 @@ frame_id_eq (struct frame_id l, struct frame_id r)
     /* Like a NaN, if either ID is invalid, the result is false.
        Note that a frame ID is invalid iff it is the null frame ID.  */
     eq = 0;
-  else if (l.stack_addr != r.stack_addr)
+  else if ((l.stack_addr != r.stack_addr && l.stack_addr > max_addr
+	   && r.stack_addr > max_addr)
+	   || (l.stack_addr & max_addr) != (r.stack_addr & max_addr))

[eaeltsin, 2013/08/06 18:14:42]

This definitely needs a comment, explaining what this does and also marking this
as a temporary, non-upstreamable thing.

What is the origin of l.stack_addr and r.stack_addr. If these are both extracted
from stack, then both should probably be 32-bit or 64-bit simultaneously? How
rbp value gets into these variables?

BTW, seems the following works OK in both modes:
  (l.stack_addr & ADDR_MASK == r.stack_addr & ADDR_MASK)

[halyavin, 2013/08/07 08:04:25]

One value is read from the stack, the other is read from the register.
Yes, but it increase false positives for normal 64-bit mode.

     /* If .stack addresses are different, the frames are different.  */
     eq = 0;
   else if (l.code_addr_p && r.code_addr_p && l.code_addr != r.code_addr)