PATCH, rs6000: silence warnings

PATCH, rs6000: silence warnings

[Ben Elliston]

When building libgcc on powerpc64-linux, I see:

/home/bje/source/gcc-trunk/libgcc/../gcc/config/rs6000/linux-unwind.h:98: warning: cast discards qualifiers from pointer target type
/home/bje/source/gcc-trunk/libgcc/../gcc/config/rs6000/linux-unwind.h:99: warning: cast discards qualifiers from pointer target type
/home/bje/source/gcc-trunk/libgcc/../gcc/config/rs6000/linux-unwind.h:101: warning: cast discards qualifiers from pointer target type
/home/bje/source/gcc-trunk/libgcc/../gcc/config/rs6000/linux-unwind.h:110: warning: cast discards qualifiers from pointer target type
/home/bje/source/gcc-trunk/libgcc/../gcc/config/rs6000/linux-unwind.h:120: warning: cast discards qualifiers from pointer target type

This patch cleans up these casts by eliminating them altogether.  With
this change, the code becomes more typesafe and easier to read.  OK for
mainline?  (Regression test and bootstrap underway on power64-linux).

Thanks,
Ben


2009-04-21  Ben Elliston  <bje@au.ibm.com>

        * config/rs6000/linux-unwind.h (get_regs): Remove type
        puns. Change the type of `pc' to an array of unsigned ints and
        update all users.  Constify frame24.

Index: /home/bje/source/gcc-trunk/gcc/config/rs6000/linux-unwind.h
===================================================================
--- /home/bje/source/gcc-trunk/gcc/config/rs6000/linux-unwind.h (revision 146466)
+++ /home/bje/source/gcc-trunk/gcc/config/rs6000/linux-unwind.h (working copy)
@@ -91,14 +91,13 @@
 static struct gcc_regs *
 get_regs (struct _Unwind_Context *context)
 {
-  const unsigned char *pc = context->ra;
+  const unsigned int *pc = context->ra;
 
   /* addi r1, r1, 128; li r0, 0x0077; sc  (sigreturn) */
   /* addi r1, r1, 128; li r0, 0x00AC; sc  (rt_sigreturn) */
-  if (*(unsigned int *) (pc + 0) != 0x38210000 + SIGNAL_FRAMESIZE
-      || *(unsigned int *) (pc + 8) != 0x44000002)
+  if (pc[0] != 0x38210000 + SIGNAL_FRAMESIZE || pc[2] != 0x44000002)
     return NULL;
-  if (*(unsigned int *) (pc + 4) == 0x38000077)
+  if (pc[1] == 0x38000077)
     {
       struct sigframe {
        char gap[SIGNAL_FRAMESIZE];
@@ -107,17 +106,17 @@
       } *frame = (struct sigframe *) context->cfa;
       return frame->regs;
     }
-  else if (*(unsigned int *) (pc + 4) == 0x380000AC)
+  else if (pc[1] == 0x380000AC)
     {
       /* This works for 2.4 kernels, but not for 2.6 kernels with vdso
         because pc isn't pointing into the stack.  Can be removed when
         no one is running 2.4.19 or 2.4.20, the first two ppc64
         kernels released.  */
-      struct rt_sigframe_24 {
+      const struct rt_sigframe_24 {
        int tramp[6];
        void *pinfo;
        struct gcc_ucontext *puc;
-      } *frame24 = (struct rt_sigframe_24 *) pc;
+      } *frame24 = (const struct rt_sigframe_24 *) context->ra;
 
       /* Test for magic value in *puc of vdso.  */
       if ((long) frame24->puc != -21 * 8)
@@ -146,16 +145,15 @@
 static struct gcc_regs *
 get_regs (struct _Unwind_Context *context)
 {
-  const unsigned char *pc = context->ra;
+  const unsigned int *pc = context->ra;
 
   /* li r0, 0x7777; sc  (sigreturn old)  */
   /* li r0, 0x0077; sc  (sigreturn new)  */
   /* li r0, 0x6666; sc  (rt_sigreturn old)  */
   /* li r0, 0x00AC; sc  (rt_sigreturn new)  */
-  if (*(const unsigned int *) (pc + 4) != 0x44000002)
+  if (pc[1] != 0x44000002)
     return NULL;
-  if (*(const unsigned int *) (pc + 0) == 0x38007777
-      || *(const unsigned int *) (pc + 0) == 0x38000077)
+  if (pc[0] == 0x38007777 || pc[0] == 0x38000077)
     {
       struct sigframe {
        char gap[SIGNAL_FRAMESIZE];
@@ -164,8 +162,7 @@
       } *frame = (struct sigframe *) context->cfa;
       return frame->regs;
     }
-  else if (*(const unsigned int *) (pc + 0) == 0x38006666
-          || *(const unsigned int *) (pc + 0) == 0x380000AC)
+  else if (pc[0] == 0x38006666 || pc[0] == 0x380000AC)
     {
       struct rt_sigframe {
        char gap[SIGNAL_FRAMESIZE + 16];

Re: PATCH, rs6000: silence warnings

[David Edelsohn]

On Tue, Apr 21, 2009 at 7:48 PM, Ben Elliston <bje@au1.ibm.com> wrote:
> On Mon, 2009-04-20 at 22:53 -0400, David Edelsohn wrote:
>
>> > This patch cleans up these casts by eliminating them altogether.  With
>> > this change, the code becomes more typesafe and easier to read.  OK for
>> > mainline?  (Regression test and bootstrap underway on power64-linux).
>>
>> Why was it initially written as char * ?  Normal, valid PowerPC code
>> should be word-aligned, but is it possible to be called with an
>> unaligned pointer?  Will this change cause problems on embedded
>> processors with strict alignment requirements?
>
> My reading was that the char * was only the type for a buffer where
> several addresses were stashed.  That does not dictate anything about
> the alignment of the addresses stored therein.

Not the alignment of the addresses, but the alignment of the buffer
itself.  It looks like it always will be suitably aligned.

The patch is okay.

Thanks, David