[PATCH] Move ELF program header

[PATCH] Move ELF program header

[Bean]

[-- Attachment #1: Type: text/plain, Size: 115 bytes --]

Hi,

This patch move the elf program header to the beginning, it should
solve the problem with coreboot.

-- 
Bean

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: elfhdr.diff --]
[-- Type: text/x-diff; name=elfhdr.diff, Size: 2519 bytes --]

diff --git a/util/elf/grub-mkimage.c b/util/elf/grub-mkimage.c
index ca138dd..51c4e54 100644
--- a/util/elf/grub-mkimage.c
+++ b/util/elf/grub-mkimage.c
@@ -166,8 +166,8 @@ add_segments (char *dir, FILE *out, int chrp, char *mods[])
   FILE *in;
   char *kernel_path;
   grub_addr_t grub_end = 0;
-  off_t phdroff;
-  int i;
+  off_t offset;
+  int i, phdr_size;
 
   /* Read ELF header.  */
   kernel_path = grub_util_get_path (dir, "kernel.elf");
@@ -177,8 +177,21 @@ add_segments (char *dir, FILE *out, int chrp, char *mods[])
 
   grub_util_read_at (&ehdr, sizeof (ehdr), 0, in);
   
-  phdrs = xmalloc (grub_target_to_host16 (ehdr.e_phentsize)
-		   * (grub_target_to_host16 (ehdr.e_phnum) + 2));
+  offset = ALIGN_UP (sizeof (ehdr), sizeof (long));
+  ehdr.e_phoff = grub_host_to_target32 (offset);
+
+  phdr_size = (grub_target_to_host16 (ehdr.e_phentsize) *
+               grub_target_to_host16 (ehdr.e_phnum));
+
+  if (mods[0] != NULL)
+    phdr_size += grub_target_to_host16 (ehdr.e_phentsize);
+
+  if (chrp)
+    phdr_size += grub_target_to_host16 (ehdr.e_phentsize);
+
+  phdrs = xmalloc (phdr_size);
+  offset += ALIGN_UP (phdr_size, sizeof (long));
+
   /* Copy all existing segments.  */
   for (i = 0; i < grub_target_to_host16 (ehdr.e_phnum); i++)
     {
@@ -207,8 +220,11 @@ add_segments (char *dir, FILE *out, int chrp, char *mods[])
   
       grub_util_read_at (segment_img, grub_target_to_host32 (phdr->p_filesz),
 			 grub_target_to_host32 (phdr->p_offset), in);
+
+      phdr->p_offset = grub_host_to_target32 (offset);
       grub_util_write_image_at (segment_img, grub_target_to_host32 (phdr->p_filesz),
-				grub_target_to_host32 (phdr->p_offset), out);
+				offset, out);
+      offset += ALIGN_UP (grub_target_to_host32 (phdr->p_filesz), sizeof (long));
 
       free (segment_img);
     }
@@ -249,14 +265,10 @@ add_segments (char *dir, FILE *out, int chrp, char *mods[])
   ehdr.e_shnum = 0;
   ehdr.e_shstrndx = 0;
 
-  /* Append entire segment table to the file.  */
-  phdroff = ALIGN_UP (grub_util_get_fp_size (out), sizeof (long));
-  grub_util_write_image_at (phdrs, grub_target_to_host16 (ehdr.e_phentsize)
-			    * grub_target_to_host16 (ehdr.e_phnum), phdroff,
-			    out);
+  /* Write entire segment table to the file.  */
+  grub_util_write_image_at (phdrs, phdr_size, grub_target_to_host32 (ehdr.e_phoff), out);
 
   /* Write ELF header.  */
-  ehdr.e_phoff = grub_host_to_target32 (phdroff);
   grub_util_write_image_at (&ehdr, sizeof (ehdr), 0, out);
 
   free (phdrs);

Re: [PATCH] Move ELF program header

[Bean]

On Mon, Jun 2, 2008 at 11:41 PM, Bean <bean123ch@gmail.com> wrote:
> Hi,
>
> This patch move the elf program header to the beginning, it should
> solve the problem with coreboot.
>
> --
> Bean
>

Any comment ?

-- 
Bean

Re: [PATCH] Move ELF program header

[Robert Millan]

On Fri, Jun 13, 2008 at 03:08:44PM +0800, Bean wrote:
> On Mon, Jun 2, 2008 at 11:41 PM, Bean <bean123ch@gmail.com> wrote:
> > Hi,
> >
> > This patch move the elf program header to the beginning, it should
> > solve the problem with coreboot.
> >
> > --
> > Bean
> >
> 
> Any comment ?

I don't have time to check atm, but please make sure it doesn't break
ieee1275 support.

-- 
Robert Millan

<GPLv2> I know my rights; I want my phone call!
<DRM> What good is a phone call… if you are unable to speak?
(as seen on /.)

Re: [PATCH] Move ELF program header

[Bean]

On Sat, Jun 14, 2008 at 7:37 PM, Robert Millan <rmh@aybabtu.com> wrote:
> On Fri, Jun 13, 2008 at 03:08:44PM +0800, Bean wrote:
>> On Mon, Jun 2, 2008 at 11:41 PM, Bean <bean123ch@gmail.com> wrote:
>> > Hi,
>> >
>> > This patch move the elf program header to the beginning, it should
>> > solve the problem with coreboot.
>> >
>> > --
>> > Bean
>> >
>>
>> Any comment ?
>
> I don't have time to check atm, but please make sure it doesn't break
> ieee1275 support.

Hi,

I test it with olpc and it works. I don't have ppc, would someone check it ?

-- 
Bean

Re: [PATCH] Move ELF program header

[Pavel Roskin]

On Sun, 2008-06-15 at 00:31 +0800, Bean wrote:

> I test it with olpc and it works. I don't have ppc, would someone check it ?

I could make CVS GRUB compile on PowerPC after removing floating point
code, but I could not get the rescue image to boot in qemu, with or
without the patch.  I know that it worked before (it would show "out of
memory" due to Open Hack'Ware limitations, but at least the GRUB code
was loaded and running).

I'm afraid I won't be able to spend any more time on the PowerPC port.
If anyone cares, they would need to fix a lot of stuff.

-- 
Regards,
Pavel Roskin

Re: [PATCH] Move ELF program header

[Bean]

On Mon, Jun 16, 2008 at 1:40 PM, Pavel Roskin <proski@gnu.org> wrote:
> On Sun, 2008-06-15 at 00:31 +0800, Bean wrote:
>
>> I test it with olpc and it works. I don't have ppc, would someone check it ?
>
> I could make CVS GRUB compile on PowerPC after removing floating point
> code, but I could not get the rescue image to boot in qemu, with or
> without the patch.  I know that it worked before (it would show "out of
> memory" due to Open Hack'Ware limitations, but at least the GRUB code
> was loaded and running).
>
> I'm afraid I won't be able to spend any more time on the PowerPC port.
> If anyone cares, they would need to fix a lot of stuff.

Hi,

I see, you're using qemu to test ppc, maybe I can try it some time. Is
the emulation good, how close to the real hardware ?

btw, do you remember which version works ? We can track down the
problem from there.

-- 
Bean

Re: [PATCH] Move ELF program header

[Pavel Roskin]

Quoting Bean <bean123ch@gmail.com>:

> I see, you're using qemu to test ppc, maybe I can try it some time. Is
> the emulation good, how close to the real hardware ?

qemu uses so called Open Hack'Ware, which doesn't have proper terminal  
support for the menu and doesn't report available memory in the way  
GRUB expects.  The later causes "out of memory" messages, but it can  
be worked around in GRUB by hardcoding memory size.

> btw, do you remember which version works ? We can track down the
> problem from there.

It was working when grub-mkrescue was added for PowerPC.  According to  
the logs, that's January 31, 2008.

It looks like my G3 PowerMac died yesterday and won't turn on anymore.  
  The same thing happened to my G4 PowerMac several months ago.  So, I  
guess qemu remains the only option for me.

-- 
Regards,
Pavel Roskin

Re: [PATCH] Move ELF program header

[Bean]

On Mon, Jun 16, 2008 at 10:33 PM, Pavel Roskin <proski@gnu.org> wrote:
> Quoting Bean <bean123ch@gmail.com>:
>
>> I see, you're using qemu to test ppc, maybe I can try it some time. Is
>> the emulation good, how close to the real hardware ?
>
> qemu uses so called Open Hack'Ware, which doesn't have proper terminal
> support for the menu and doesn't report available memory in the way GRUB
> expects.  The later causes "out of memory" messages, but it can be worked
> around in GRUB by hardcoding memory size.
>
>> btw, do you remember which version works ? We can track down the
>> problem from there.
>
> It was working when grub-mkrescue was added for PowerPC.  According to the
> logs, that's January 31, 2008.
>
> It looks like my G3 PowerMac died yesterday and won't turn on anymore.  The
> same thing happened to my G4 PowerMac several months ago.  So, I guess qemu
> remains the only option for me.

Hi,

Oh, this problem is caused by grub_ofconsole_setcursor. Open Hackware
doesn't like the cursor-on call, if I comment it out, it will continue
to run until the "out of memory" error.

-- 
Bean

Re: [PATCH] Move ELF program header

[Robert Millan]

On Tue, Jun 17, 2008 at 10:21:43PM +0800, Bean wrote:
> 
> Hi,
> 
> Oh, this problem is caused by grub_ofconsole_setcursor. Open Hackware
> doesn't like the cursor-on call, if I comment it out, it will continue
> to run until the "out of memory" error.

We could have a flag in kern/ieee1275/cmain.c for this.  There's a stanza for
Open Hackware already.

-- 
Robert Millan

<GPLv2> I know my rights; I want my phone call!
<DRM> What good is a phone call… if you are unable to speak?
(as seen on /.)

Re: [PATCH] Move ELF program header

[Bean]

On Tue, Jun 17, 2008 at 11:17 PM, Robert Millan <rmh@aybabtu.com> wrote:
> On Tue, Jun 17, 2008 at 10:21:43PM +0800, Bean wrote:
>>
>> Hi,
>>
>> Oh, this problem is caused by grub_ofconsole_setcursor. Open Hackware
>> doesn't like the cursor-on call, if I comment it out, it will continue
>> to run until the "out of memory" error.
>
> We could have a flag in kern/ieee1275/cmain.c for this.  There's a stanza for
> Open Hackware already.

Hi,

That flag is not enough, actually, Open Hackware don't support
interpret command at all, perhaps we need to a new flag
GRUB_IEEE1275_FLAG_CANNOT_INTERPRET.

-- 
Bean

Re: [PATCH] Move ELF program header

[Pavel Roskin]

On Tue, 2008-06-17 at 17:17 +0200, Robert Millan wrote:
> On Tue, Jun 17, 2008 at 10:21:43PM +0800, Bean wrote:
> > 
> > Hi,
> > 
> > Oh, this problem is caused by grub_ofconsole_setcursor. Open Hackware
> > doesn't like the cursor-on call, if I comment it out, it will continue
> > to run until the "out of memory" error.
> 
> We could have a flag in kern/ieee1275/cmain.c for this.  There's a stanza for
> Open Hackware already.

Actually, the real OpenFirmware on PowerMac G3 is also complaining about
"cursor on", but it's not fatal.  Open Hackware doesn't have Forth
interpreter, so it recognizes Forth commands by checksums and panics for
unknown checksums.

I could resurrect my Mac by replugging everything on the motherboard.

The real OpenFirmware fails to recognize modules embedded into the image
by grub-mkimage.  I remember I spent some time fixing it, but it's
broken again.

-- 
Regards,
Pavel Roskin

Re: [PATCH] Move ELF program header

[Robert Millan]

On Tue, Jun 17, 2008 at 11:32:15AM -0400, Pavel Roskin wrote:
> 
> The real OpenFirmware fails to recognize modules embedded into the image
> by grub-mkimage.  I remember I spent some time fixing it, but it's
> broken again.

That's strange, it does work on i386 with the "real" OpenFirmware (I think
you're referring to the same implementation)

-- 
Robert Millan

<GPLv2> I know my rights; I want my phone call!
<DRM> What good is a phone call… if you are unable to speak?
(as seen on /.)

Re: [PATCH] Move ELF program header

[Pavel Roskin]

On Tue, 2008-06-17 at 17:34 +0200, Robert Millan wrote:
> On Tue, Jun 17, 2008 at 11:32:15AM -0400, Pavel Roskin wrote:
> > 
> > The real OpenFirmware fails to recognize modules embedded into the image
> > by grub-mkimage.  I remember I spent some time fixing it, but it's
> > broken again.
> 
> That's strange, it does work on i386 with the "real" OpenFirmware (I think
> you're referring to the same implementation)

I don't know if it's the same implementation.  It's whatever Apple put
on their G3 machines.

-- 
Regards,
Pavel Roskin

Re: [PATCH] Move ELF program header

[Bean]

On Tue, Jun 17, 2008 at 11:32 PM, Pavel Roskin <proski@gnu.org> wrote:
> The real OpenFirmware fails to recognize modules embedded into the image
> by grub-mkimage.  I remember I spent some time fixing it, but it's
> broken again.

Hi,

Are you using the native compiler, or cross compile from i386 ? It
could have some effect.

-- 
Bean

Re: [PATCH] Move ELF program header

[Pavel Roskin]

On Tue, 2008-06-17 at 23:47 +0800, Bean wrote:
> On Tue, Jun 17, 2008 at 11:32 PM, Pavel Roskin <proski@gnu.org> wrote:
> > The real OpenFirmware fails to recognize modules embedded into the image
> > by grub-mkimage.  I remember I spent some time fixing it, but it's
> > broken again.
> 
> Hi,
> 
> Are you using the native compiler, or cross compile from i386 ? It
> could have some effect.

I'm compiling natively on PowerPC.

-- 
Regards,
Pavel Roskin

Re: [PATCH] Move ELF program header

[Bean]

On Tue, Jun 17, 2008 at 11:55 PM, Pavel Roskin <proski@gnu.org> wrote:
> On Tue, 2008-06-17 at 23:47 +0800, Bean wrote:
>> On Tue, Jun 17, 2008 at 11:32 PM, Pavel Roskin <proski@gnu.org> wrote:
>> > The real OpenFirmware fails to recognize modules embedded into the image
>> > by grub-mkimage.  I remember I spent some time fixing it, but it's
>> > broken again.
>>
>> Hi,
>>
>> Are you using the native compiler, or cross compile from i386 ? It
>> could have some effect.
>
> I'm compiling natively on PowerPC.

It could be an endian problem. Check the elf header, see if the fields are ok.

-- 
Bean

Re: [PATCH] Move ELF program header

[Pavel Roskin]

On Wed, 2008-06-18 at 00:02 +0800, Bean wrote:
> >> Are you using the native compiler, or cross compile from i386 ? It
> >> could have some effect.
> >
> > I'm compiling natively on PowerPC.
> 
> It could be an endian problem. Check the elf header, see if the fields are ok.

That's the headers, I don't see anything wrong.  The sections are
missing, but it has always been like that.  The version is current from
CVS.

$ ./grub-mkimage -o grub.img -d . *.mod
$ objdump -x grub.img

grub.img:     file format elf32-powerpc
grub.img
architecture: powerpc:common, flags 0x00000102:
EXEC_P, D_PAGED
start address 0x00010000

Program Header:
    LOAD off    0x000000c0 vaddr 0x00010000 paddr 0x00010000 align 2**4
         filesz 0x0000bf04 memsz 0x000133a8 flags rwx
    LOAD off    0x0000bfc4 vaddr 0x100000b4 paddr 0x100000b4 align 2**2
         filesz 0x00000024 memsz 0x00000024 flags r--
    NOTE off    0x0000bfc4 vaddr 0x100000b4 paddr 0x100000b4 align 2**2
         filesz 0x00000024 memsz 0x00000024 flags r--
   STACK off    0x00000000 vaddr 0x00000000 paddr 0x00000000 align 2**2
         filesz 0x00000000 memsz 0x00000000 flags rwx
    LOAD off    0x0000bfe8 vaddr 0x10009000 paddr 0x10009000 align 2**2
         filesz 0x00052104 memsz 0x00052104 flags rwx

Sections:
Idx Name          Size      VMA       LMA       File off  Algn
SYMBOL TABLE:
no symbols


The image loads, complains about "cursor-on", shows "Welcome to GRUB"
and the prompt.  "lsmod" shows an empty list.

-- 
Regards,
Pavel Roskin

Re: [PATCH] Move ELF program header

[Bean]

On Wed, Jun 18, 2008 at 1:18 AM, Pavel Roskin <proski@gnu.org> wrote:
> On Wed, 2008-06-18 at 00:02 +0800, Bean wrote:
>> >> Are you using the native compiler, or cross compile from i386 ? It
>> >> could have some effect.
>> >
>> > I'm compiling natively on PowerPC.
>>
>> It could be an endian problem. Check the elf header, see if the fields are ok.
>
> That's the headers, I don't see anything wrong.  The sections are
> missing, but it has always been like that.  The version is current from
> CVS.
>
> $ ./grub-mkimage -o grub.img -d . *.mod
> $ objdump -x grub.img
>
> grub.img:     file format elf32-powerpc
> grub.img
> architecture: powerpc:common, flags 0x00000102:
> EXEC_P, D_PAGED
> start address 0x00010000
>
> Program Header:
>    LOAD off    0x000000c0 vaddr 0x00010000 paddr 0x00010000 align 2**4
>         filesz 0x0000bf04 memsz 0x000133a8 flags rwx
>    LOAD off    0x0000bfc4 vaddr 0x100000b4 paddr 0x100000b4 align 2**2
>         filesz 0x00000024 memsz 0x00000024 flags r--
>    NOTE off    0x0000bfc4 vaddr 0x100000b4 paddr 0x100000b4 align 2**2
>         filesz 0x00000024 memsz 0x00000024 flags r--
>   STACK off    0x00000000 vaddr 0x00000000 paddr 0x00000000 align 2**2
>         filesz 0x00000000 memsz 0x00000000 flags rwx
>    LOAD off    0x0000bfe8 vaddr 0x10009000 paddr 0x10009000 align 2**2
>         filesz 0x00052104 memsz 0x00052104 flags rwx
>
> Sections:
> Idx Name          Size      VMA       LMA       File off  Algn
> SYMBOL TABLE:
> no symbols
>
>
> The image loads, complains about "cursor-on", shows "Welcome to GRUB"
> and the prompt.  "lsmod" shows an empty list.

Hi,

You can also add some printf in kern/elf.c to see if grub can parse it
properly, for example, grub_elf_check_header is a common place where
problem occurs.

-- 
Bean