Re: OpenRISC arch/platform fixes

[Scott Furman <sfurman at rosum dot com>]

[Away from email for awhile.  Sorry for not responding sooner...]

Jonathan Larmour wrote:

> > + + /* FIXME: libsupc++ is not needed by eCos, but it might be needed 
> > by apps linked w/ eCos.
> > +    However, it does not yet compile for OpenRISC. */
>
>
> That's unfortunate. Normally at least libsupc++ compiles, even if not 
> full libstdc++. 

It may well be easy to fix.  (I haven't invested any time in it.)  If 
someone else can figure it out, please let me know and we can add 
libsupc++ back to the linker script.

> > ***************
> > *** 270,276 ****
> >       {                                                         \
> >       __SDATA_START__ = ABSOLUTE(.);                            \
> >         FORCE_OUTPUT; *(.sdata) *(.sdata.*) *(.gnu.linkonce.s*) \
> > !     } > _region_
> >     #define SECTION_lit8(_region_, _vma_, _lma_)    \
> >     .lit8 _vma_ : _lma_                           \
> > --- 273,281 ----
> >       {                                                         \
> >       __SDATA_START__ = ABSOLUTE(.);                            \
> >         FORCE_OUTPUT; *(.sdata) *(.sdata.*) *(.gnu.linkonce.s*) \
> > !     } > _region_                                              \
> > !     __ram_data_end = .; _edata = . ;                          \
> > !     PROVIDE (edata = .);
> >     #define SECTION_lit8(_region_, _vma_, _lma_)    \
> >     .lit8 _vma_ : _lma_                           \
> > ***************
> > *** 283,290 ****
> >       {                                           \
> >         FORCE_OUTPUT; *(.lit4)                    \
> >       } > _region_                                \
> > -   __ram_data_end = .; _edata = . ;              \
> > -   PROVIDE (edata = .);
> >     #define SECTION_sbss(_region_, _vma_, _lma_)                     \
> >       .sbss _vma_ : 
> > _lma_                                                         \
>
>
> If C symbols have underscore prepended, you may want to change _edata 
> to __edata to prevent possible namespace pollution (yes I know there 
> are a few other examples already in the HALs and given time we'd get 
> to those). 

The attached patch changes _edata =>__edata and _end =>__end.

> I'm slightly curious as to why you moved these lines from the bottom 
> of lit4 to the bottom of sdata. The .lit sections (if they exist) are 
> also initialized data so would need relocating too. 

By "relocating", I assume that you're referring to copying from ROM to 
RAM, and not talking about some linker operation.  I believe that the 
lit4 and lit8 sections contain only shared literals and are, hence, 
read-only.  My thinking was that they would be retained in ROM (along 
with .rdonly, .ctors, and other read-only sections) with a ROM or ROMRAM 
configuration.

Incidentally, lit4 and lit8 aren't currently used for the OpenRISC 
architecture.

> > Index: orp/current/include/pkgconf/mlt_openrisc_orp_ram.ldi
> > ===================================================================
> > RCS file: 
> > /export/CVS/dave/ecos/packages/hal/openrisc/orp/current/include/pkgconf/mlt_openrisc_orp_ram.ldi,v
> > retrieving revision 1.1.1.1
> > diff -s -c -r1.1.1.1 mlt_openrisc_orp_ram.ldi
> > *** orp/current/include/pkgconf/mlt_openrisc_orp_ram.ldi    22 May 
> > 2003 00:37:33 -0000    1.1.1.1
> > --- orp/current/include/pkgconf/mlt_openrisc_orp_ram.ldi    18 Jun 
> > 2003 02:55:53 -0000
> > ***************
> > *** 11,38 ****
> >   SECTIONS
> >   {
> >       SECTIONS_BEGIN
> > !     SECTION_vectors (ram, 0x00000100, LMA_EQ_VMA)
> >       SECTION_ROMISC (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_RELOCS (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_init (ram, ALIGN (0x8), LMA_EQ_VMA)
> > -     SECTION_text (ram, 0x00008000, LMA_EQ_VMA)
> >       SECTION_fini (ram, ALIGN (0x4), LMA_EQ_VMA)
> >       SECTION_rodata1 (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_rodata (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_fixup (ram, ALIGN (0x4), LMA_EQ_VMA)
> >       SECTION_gcc_except_table (ram, ALIGN (0x4), LMA_EQ_VMA)
> > -     SECTION_data (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_eh_frame (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_ctors (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_dtors (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_devtab (ram, ALIGN (0x8), LMA_EQ_VMA)
> > -     SECTION_got (ram, ALIGN (0x8), LMA_EQ_VMA)
> > -     SECTION_dynamic (ram, ALIGN (0x8), LMA_EQ_VMA)
> > -     SECTION_sdata (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_lit8 (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_lit4 (ram, ALIGN (0x8), LMA_EQ_VMA)
> > !     SECTION_sbss (ram, ALIGN (0x8), LMA_EQ_VMA)
> > !     SECTION_bss (ram, ALIGN (0x10), LMA_EQ_VMA)
> >       CYG_LABEL_DEFN(__heap1) = ALIGN (0x8);
> >       SECTIONS_END
> >   }
> > --- 11,40 ----
> >   SECTIONS
> >   {
> >       SECTIONS_BEGIN
> > !     SECTION_vectors (ram, 0x00000000, LMA_EQ_VMA)
> > !     SECTION_text (ram, 0x00008000, LMA_EQ_VMA)
> > ! !     SECTION_data (ram, ALIGN(0x8), LMA_EQ_VMA)
> > !     SECTION_got (ram, ALIGN (0x8), LMA_EQ_VMA)
> > !     SECTION_dynamic (ram, ALIGN (0x8), LMA_EQ_VMA)
> > !     SECTION_sdata (ram, ALIGN (0x8), LMA_EQ_VMA)
> > !     SECTION_sbss (ram, ALIGN (0x8), LMA_EQ_VMA)
> > !     SECTION_bss (ram, ALIGN (0x10), LMA_EQ_VMA)
> >       SECTION_ROMISC (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_RELOCS (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_init (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_fini (ram, ALIGN (0x4), LMA_EQ_VMA)
> >       SECTION_rodata1 (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_rodata (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_fixup (ram, ALIGN (0x4), LMA_EQ_VMA)
> >       SECTION_gcc_except_table (ram, ALIGN (0x4), LMA_EQ_VMA)
> >       SECTION_eh_frame (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_ctors (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_dtors (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_devtab (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_lit8 (ram, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_lit4 (ram, ALIGN (0x8), LMA_EQ_VMA)
> > !       CYG_LABEL_DEFN(__heap1) = ALIGN (0x8);
> >       SECTIONS_END
> >   }
>
>
> This is asking for trouble with future toolchains, or with C++ 
> exception support down the line. If the sections are empty the order 
> won't matter. If they're not empty, it's probably for a good reason. 

When you say "this is asking for trouble", I assume you're referring to 
the re-ordering of sections.  For the ROM linker script, I had to 
re-order the sections to aggregate those (read-only) sections which 
could remain in ROM versus those sections that were mutable and, hence, 
had to be copied to RAM.  I re-used the resulting section ordering in 
the *_ram.ldi file.  I can revert the section reordering for the 
*_ram.ldi file, but what should I do about the *_rom.ldi file ?  (I 
would prefer to keep them similar, if possible).

> > Index: orp/current/include/pkgconf/mlt_openrisc_orp_rom.ldi
> > ===================================================================
> > RCS file: 
> > /export/CVS/dave/ecos/packages/hal/openrisc/orp/current/include/pkgconf/mlt_openrisc_orp_rom.ldi,v
> > retrieving revision 1.1.1.1
> > diff -s -c -r1.1.1.1 mlt_openrisc_orp_rom.ldi
> > *** orp/current/include/pkgconf/mlt_openrisc_orp_rom.ldi    22 May 
> > 2003 00:37:33 -0000    1.1.1.1
> > --- orp/current/include/pkgconf/mlt_openrisc_orp_rom.ldi    18 Jun 
> > 2003 03:00:41 -0000
> > ***************
> > *** 11,17 ****
> >   SECTIONS
> >   {
> >       SECTIONS_BEGIN
> > !     SECTION_vectors (rom, 0xF0000100, LMA_EQ_VMA)
> >       SECTION_ROMISC (rom, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_RELOCS (rom, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_init (rom, ALIGN (0x8), LMA_EQ_VMA)
> > --- 11,31 ----
> >   SECTIONS
> >   {
> >       SECTIONS_BEGIN
> > ! !     SECTION_vectors (rom, 0xF0000000, LMA_EQ_VMA)
> > ! !     // All eCos code is put in the .text section unless 
> > explicitly placed
> > !     // into a section named section .text.ram*.
> > !     SECTION_text_ram (ram, 0x00001000, AT(0xF0001000))
> > !     SECTION_data (ram, ALIGN(0x8), FOLLOWING(.text.ram))
> > !     SECTION_got (ram, ALIGN (0x8), FOLLOWING (.data))
> > !     SECTION_dynamic (ram, ALIGN (0x8), FOLLOWING (.got))
> > !     SECTION_sdata (ram, ALIGN (0x8), FOLLOWING (.dynamic))
> > !     SECTION_sbss (ram, ALIGN (0x8), LMA_EQ_VMA)
> > !     SECTION_bss (ram, ALIGN (0x10), LMA_EQ_VMA)
> > !     CYG_LABEL_DEFN(__heap1) = ALIGN (0x8);
> > ! !     . = LOADADDR(.sdata) + SIZEOF(.sdata);
> >       SECTION_ROMISC (rom, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_RELOCS (rom, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_init (rom, ALIGN (0x8), LMA_EQ_VMA)
> > ***************
> > *** 21,38 ****
> >       SECTION_rodata (rom, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_fixup (rom, ALIGN (0x4), LMA_EQ_VMA)
> >       SECTION_gcc_except_table (rom, ALIGN (0x4), LMA_EQ_VMA)
> > !     SECTION_data (ram, 0x1000, FOLLOWING (.gcc_except_table))
> > !     SECTION_eh_frame (ram, ALIGN (0x8), FOLLOWING (.data))
> > !     SECTION_ctors (ram, ALIGN (0x8), FOLLOWING (.eh_frame))
> > !     SECTION_dtors (ram, ALIGN (0x8), FOLLOWING (.ctors))
> > !     SECTION_devtab (ram, ALIGN (0x8), FOLLOWING (.dtors))
> > !     SECTION_got (ram, ALIGN (0x8), FOLLOWING (.devtab))
> > !     SECTION_dynamic (ram, ALIGN (0x8), FOLLOWING (.got))
> > !     SECTION_sdata (ram, ALIGN (0x8), FOLLOWING (.dynamic))
> > !     SECTION_lit8 (ram, ALIGN (0x8), FOLLOWING (.sdata))
> > !     SECTION_lit4 (ram, ALIGN (0x8), FOLLOWING (.lit8))
> > !     SECTION_sbss (ram, ALIGN (0x8), LMA_EQ_VMA)
> > !     SECTION_bss (ram, ALIGN (0x10), LMA_EQ_VMA)
> > !     CYG_LABEL_DEFN(__heap1) = ALIGN (0x8);
> >       SECTIONS_END
> >   }
> > --- 35,46 ----
> >       SECTION_rodata (rom, ALIGN (0x8), LMA_EQ_VMA)
> >       SECTION_fixup (rom, ALIGN (0x4), LMA_EQ_VMA)
> >       SECTION_gcc_except_table (rom, ALIGN (0x4), LMA_EQ_VMA)
> > !     SECTION_eh_frame (rom, ALIGN (0x8), LMA_EQ_VMA)
> > !     SECTION_ctors (rom, ALIGN (0x8), LMA_EQ_VMA)
> > !     SECTION_dtors (rom, ALIGN (0x8), LMA_EQ_VMA)
> > !     SECTION_devtab (rom, ALIGN (0x8), LMA_EQ_VMA)
> > !     SECTION_lit8 (rom, ALIGN (0x8), LMA_EQ_VMA)
> > !     SECTION_lit4 (rom, ALIGN (0x8), LMA_EQ_VMA)
> > !       SECTIONS_END
> >   }
>
>
> The same applies here with the ordering, although the change to rom is 
> not unreasonable for all of these, with the exception of the .lit 
> sections which must be relocated to ram like .data.
>
> There's some outstanding issues from before:
>
> * The invocation of cyg_hal_exception_handler doesn't look right... 
> nothing is saved! Otherwise if you take an exception there's no saved 
> register state to point at... vital for the GDB stub at least. This 
> way CYGSEM_HAL_ROM_MONITOR will get used as it's meant to as well 
> (since it isn't used at all right now ;)). 

I answered this in some previous message sent to the list 4/28/03:

Scott Furman wrote:  

> All the exception state *is* stored before cyg_hal_exception_handler() 
> is called. (Check out the expansion of the exception_vector assembly 
> macro in vectors.S. ) I know that this "works" because I have hit gdb 
> exceptions for illegal instructions, bus errors, etc. and stack 
> backtraces were sensible-looking.


> * If you're using this with RedBoot do you have a potted .ecm file to 
> go with this? And also when we're done, remember to send me some 
> prebuilt known good redboot images I can check in to our redboot image 
> repository (ecos/images which isn't checked out by default so you 
> won't have seen it). 

I mentioned this last time: I'm not really clear on the process of 
creating a *.ecm file and how it differs from a *.ecc file.  (Where is 
this documented ?) 

If I use configtool and select template "ORP" and configuration 
"Redboot", I get a perfectly useable Redboot configuration.  How can I 
take the huge *.ecc file generated and turn it into a *.ecm file ?

> * I've also got to make the changes to ecos.db, and NEWS. But I also 
> need info from you to get things at 
> http://ecos.sourceware.org/hardware.html, particularly any board 
> availability information (or is it simulator only?). I can also put in 
> sponsorship info about Rosum there.

My understanding is that the prototyping boards are no longer available 
for purchase.  My guess is that most developers will not use an 
off-the-shelf board, but will instead use a custom HW platform with a 
customized version of the OpenRISC (and will define their own eCos 
platform).  Initially, though most will want to run on the simulator if 
they're not familiar with eCos.

> * Is there any way to get eCos listed as a project at 
> http://www.opencores.org/projects/or1k/Home with the other OS's? 

Once these most recent changes are checked into the repository, I will 
pester the OpenCores folks to add eCos to their list of projects.  I 
think they would be pleased to do so...

> * Some sort of documentation would be good just so people can find out 
> how to get going. Docbook would be good, but if nothing else a 
> README.txt in one or both packages might be something. 

There *is* an (admittedly very sparse) README.html file in the 
arch/current/doc subdir.  I created bug 1000030 to reflect that 
additional introductory documentation needs to be written.

-Scott

Index: src/openrisc.ld
===================================================================
RCS file: /export/CVS/dave/ecos/packages/hal/openrisc/arch/current/src/openrisc.ld,v
retrieving revision 1.4
diff -c -r1.4 openrisc.ld
*** src/openrisc.ld	16 May 2003 19:10:57 -0000	1.4
--- src/openrisc.ld	2 Jul 2003 21:02:11 -0000
***************
*** 274,280 ****
      __SDATA_START__ = ABSOLUTE(.);                            \
        FORCE_OUTPUT; *(.sdata) *(.sdata.*) *(.gnu.linkonce.s*) \
      } > _region_                                              \
!     __ram_data_end = .; _edata = . ;                          \
      PROVIDE (edata = .);
  
  #define SECTION_lit8(_region_, _vma_, _lma_)    \
--- 274,280 ----
      __SDATA_START__ = ABSOLUTE(.);                            \
        FORCE_OUTPUT; *(.sdata) *(.sdata.*) *(.gnu.linkonce.s*) \
      } > _region_                                              \
!     __ram_data_end = .; __edata = . ;                         \
      PROVIDE (edata = .);
  
  #define SECTION_lit8(_region_, _vma_, _lma_)    \
***************
*** 305,311 ****
        __bss_end = ABSOLUTE (.); }                           \
      > _region_
  
! #define SECTIONS_END . = ALIGN(4); _end = .; PROVIDE (end = .);
  
  #include <pkgconf/system.h>
  #include CYGHWR_MEMORY_LAYOUT_LDI
--- 305,311 ----
        __bss_end = ABSOLUTE (.); }                           \
      > _region_
  
! #define SECTIONS_END . = ALIGN(4); __end = .; PROVIDE (end = .);
  
  #include <pkgconf/system.h>
  #include CYGHWR_MEMORY_LAYOUT_LDI