Re: Problem with loading 64-bit ELF image with multiboot2

[Pawel Wojtalczyk]

Hello Vladimir,

I just do not understand current implementation and would not discuss as 
a bug, because you rejected this issue as a bug.

But I see that you have modified the code to allow loading 64-bit ELFs 
linked in high address and did not received the notification about it.

Thank you.

Regards
Pawel


On 22.07.2013 14:28, Vladimir 'φ-coder/phcoder' Serbinenko wrote:
> On 22.07.2013 13:46, Pawel Wojtalczyk wrote:
> > Hello,
> >
> > I use x86 64-bit GRUB2 under EFI and would like to load 64-bit ELF
> > image. The image looks as following:
> Rather than bringing the issue once more time in a different place, 
> look that this was already fixed in trunk.
> > $ objdump -h Image
> >
> > Image: file format elf64-x86-64
> >
> > Sections:
> > Idx Name Size VMA LMA File
> > off Algn
> > 0 .text 0019d870 ffffffff90008000 0000000010008000
> > 00000120 2**5
> > CONTENTS, ALLOC, LOAD, CODE
> > 1 .wrs_build_vars 00000150 ffffffff901a5870 00000000101a5870
> > 0019d990 2**0
> > CONTENTS, ALLOC, LOAD, READONLY, DATA
> > 2 .data 0000db50 ffffffff901a59c0 00000000101a59c0
> > 0019db00 2**6
> > CONTENTS, ALLOC, LOAD, DATA
> > 3 .bss 01016e40 ffffffff901b3520 00000000101b3520
> > 001ab660 2**5
> > ALLOC
> > $ objdump -f Image
> >
> > Image: file format elf64-x86-64
> > architecture: i386:x86-64, flags 0x00000012:
> > EXEC_P, HAS_SYMS
> > start address 0xffffffff90008000
> >
> > I tried to load it via GRUB2, but unfortunately it displays 'invalid
> > entry point for ELF64'. The problem is that for 64-bit ELF images the
> > multiboot_elfxx.c library still requires 32-bit virtual address:
> >
> > #ifdef MULTIBOOT_LOAD_ELF64
> > # if defined( __mips)
> > /* We still in 32-bit mode. */
> > if (ehdr->e_entry < 0xffffffff80000000ULL)
> > return grub_error (GRUB_ERR_BAD_OS, "invalid entry point for ELF64");
> > # else
> > /* We still in 32-bit mode. */
> > if (ehdr->e_entry > 0xffffffff)
> > return grub_error (GRUB_ERR_BAD_OS, "invalid entry point for ELF64");
> > # endif
> > #endif
> >
> > Could you tell me why this assumption is required? The virtual address
> > (e_entry is virtual following ELF specification) is related to the image
> > itself, how it setup MMU, not the loader (GRUB2). GRUB2 should check
> > only if the physical location of loaded image is in 32-bit physical 
> > memory.
> >
> > Besides if we skip thich e_entry check, the image is properly loaded,
> > because grub_multiboot_payload_eip is properly computed few lines below.
> >
> > grub_multiboot_payload_eip = (ehdr->e_entry - phdr(i)->p_vaddr)
> > + phdr(i)->p_paddr;
> >
> > Could you tell me why GRUB2 checks for 64-bit images this e_entry
> > virtual address (which can be everywhere in the 64-bit address space)
> > and requires it to be 32-bit address space only?
> >
> > Regards
> > Pawel Wojtalczyk
> >
> > _______________________________________________
> > Grub-devel mailing list
> > address@hidden
> > https://lists.gnu.org/mailman/listinfo/grub-devel
>
>
> _______________________________________________
> Grub-devel mailing list
> address@hidden
> https://lists.gnu.org/mailman/listinfo/grub-devel