Re: 64-bit support for srec

Re: 64-bit support for srec

[Joel Sherrill]

Lol.. who would have thought this day would come?

Can anything on the receiving end read these? 

--joel

Mike Stump <mikestump@comcast.net> wrote:

>So, I'm sure I will anger the gods, for I am a heretic, but, here is support for writing 64-bit srecs.
>
>Ok?
>

Re: 64-bit support for srec

[Mike Stump]

On Feb 1, 2012, at 4:13 PM, Joel Sherrill wrote:
> Lol.. who would have thought this day would come?

> Can anything on the receiving end read these? 

Yes, though, most existing srec readers can't handle 64-bit addresses, it goes without saying.  I do have such a reader that works just fine.

64-bit support for srec

[Mike Stump]

[-- Attachment #1: Type: text/plain, Size: 113 bytes --]

So, I'm sure I will anger the gods, for I am a heretic, but, here is support for writing 64-bit srecs.

Ok?


[-- Attachment #2: binutils-1.diffs.txt --]
[-- Type: text/plain, Size: 2549 bytes --]

2012-02-01  Mike Stump  <mikestump@comcast.net>

	* srec.c (srec_set_section_contents): Add 64-bit output support.
	(srec_write_record): Likewise.
	(srec_write_section): Likewise.	

Index: binutils/bfd/srec.c
===================================================================
*** binutils/bfd/srec.c	(revision 2070)
--- binutils/bfd/srec.c	(working copy)
*************** srec_set_section_contents (bfd *abfd,
*** 897,904 ****
        else if ((section->lma + offset + bytes_to_do - 1) <= 0xffffff
  	       && tdata->type <= 2)
  	tdata->type = 2;
!       else
  	tdata->type = 3;
  
        entry->data = data;
        entry->where = section->lma + offset;
--- 897,907 ----
        else if ((section->lma + offset + bytes_to_do - 1) <= 0xffffff
  	       && tdata->type <= 2)
  	tdata->type = 2;
!       else if ((section->lma + offset + bytes_to_do - 1) <= 0xffffffff
! 	       && tdata->type <= 2)
  	tdata->type = 3;
+       else
+ 	tdata->type = 4;
  
        entry->data = data;
        entry->where = section->lma + offset;
*************** srec_write_record (bfd *abfd,
*** 956,961 ****
--- 959,973 ----
  
    switch (type)
      {
+     case 4:
+       TOHEX (dst, (address >> 56), check_sum);
+       dst += 2;
+       TOHEX (dst, (address >> 48), check_sum);
+       dst += 2;
+       TOHEX (dst, (address >> 40), check_sum);
+       dst += 2;
+       TOHEX (dst, (address >> 32), check_sum);
+       dst += 2;
      case 3:
      case 7:
        TOHEX (dst, (address >> 24), check_sum);
*************** srec_write_section (bfd *abfd,
*** 1019,1027 ****
    /* Validate number of data bytes to write.  The srec length byte
       counts the address, data and crc bytes.  S1 (tdata->type == 1)
       records have two address bytes, S2 (tdata->type == 2) records
!      have three, and S3 (tdata->type == 3) records have four.
!      The total length can't exceed 255, and a zero data length will
!      spin for a long time.  */
    if (Chunk == 0)
      Chunk = 1;
    else if (Chunk > MAXCHUNK - tdata->type - 2)
--- 1031,1039 ----
    /* Validate number of data bytes to write.  The srec length byte
       counts the address, data and crc bytes.  S1 (tdata->type == 1)
       records have two address bytes, S2 (tdata->type == 2) records
!      have three, S3 (tdata->type == 3) records have four and S4
!      (tdata->type == 4) records have eight.  The total length can't
!      exceed 255, and a zero data length will spin for a long time.  */
    if (Chunk == 0)
      Chunk = 1;
    else if (Chunk > MAXCHUNK - tdata->type - 2)

Re: 64-bit support for srec

[Alan Modra]

On Wed, Feb 01, 2012 at 03:55:11PM -0800, Mike Stump wrote:
> --- 897,907 ----
>         else if ((section->lma + offset + bytes_to_do - 1) <= 0xffffff
>   	       && tdata->type <= 2)
>   	tdata->type = 2;
> !       else if ((section->lma + offset + bytes_to_do - 1) <= 0xffffffff
> ! 	       && tdata->type <= 2)

Probably ought to be <= 3 here.  Presumably the idea of the existing
code is that once you've used S3 records, don't drop back to S2.

> +     case 4:
> +       TOHEX (dst, (address >> 56), check_sum);
> +       dst += 2;

What if "address" is only a 32-bit type?  I think you need to
conditionalize all your new code on BFD64.  And it would be sad if we
can't read back our own output..

-- 
Alan Modra
Australia Development Lab, IBM

Re: 64-bit support for srec

[Mike Stump]

On Feb 1, 2012, at 7:22 PM, Alan Modra wrote:
> On Wed, Feb 01, 2012 at 03:55:11PM -0800, Mike Stump wrote:
>> --- 897,907 ----
>>        else if ((section->lma + offset + bytes_to_do - 1) <= 0xffffff
>>  	       && tdata->type <= 2)
>>  	tdata->type = 2;
>> !       else if ((section->lma + offset + bytes_to_do - 1) <= 0xffffffff
>> ! 	       && tdata->type <= 2)
> 
> Probably ought to be <= 3 here.  Presumably the idea of the existing
> code is that once you've used S3 records, don't drop back to S2.

Agreed:

Index: bfd/srec.c
===================================================================
--- bfd/srec.c  (revision 2081)
+++ bfd/srec.c  (working copy)
@@ -898,7 +898,7 @@ srec_set_section_contents (bfd *abfd,
               && tdata->type <= 2)
        tdata->type = 2;
       else if ((section->lma + offset + bytes_to_do - 1) <= 0xffffffff
-              && tdata->type <= 2)
+              && tdata->type <= 3)
        tdata->type = 3;
       else
        tdata->type = 4;


>> +     case 4:
>> +       TOHEX (dst, (address >> 56), check_sum);
>> +       dst += 2;
> 
> What if "address" is only a 32-bit type?

The optimizer removes the 64-bit code, since it can never be true.

> I think you need to conditionalize all your new code on BFD64.

I think it is safe to rely upon the optimizer to remove it.  I don't see any advantages to peppering the code with #ifdef.

> And it would be sad if we can't read back our own output..

It would be even sadder if we silently dropped address bits and couldn't handle 64-bit addresses...  oh, wait, we already do that.  I'll see about doing up a reader, though, I can't say if or when I will get time to do it.

Re: 64-bit support for srec

[Alan Modra]

On Thu, Feb 02, 2012 at 04:36:46PM -0800, Mike Stump wrote:
> The optimizer removes the 64-bit code, since it can never be true.

Without any warnings that cause -Werror builds to fail?

-- 
Alan Modra
Australia Development Lab, IBM

Re: 64-bit support for srec

[Mike Stump]

On Feb 2, 2012, at 5:35 PM, Alan Modra wrote:
> On Thu, Feb 02, 2012 at 04:36:46PM -0800, Mike Stump wrote:
>> The optimizer removes the 64-bit code, since it can never be true.
> 
> Without any warnings that cause -Werror builds to fail?

Yeah, I thought it might complain about that, but I tested it out with a similar test appropriate for 64-bit machines, and it didn't complain.  Love to hear if it complains for others.