I just want to report that, I do not know if it shall be considered as a bug, as bad use, as undocumented feature: Lets compile some Linux kernel with Fedora 6 default toolchain: [etienne@localhost ~]$ ld --version GNU ld version 2.17.50.0.3-6 20060715 [etienne@localhost ~]$ readelf -v GNU readelf 2.17.50.0.3-6 20060715 [etienne@localhost tmp]$ tar xjf ../linux2/linux-2.6.20.tar.bz2 [etienne@localhost tmp]$ cd linux-2.6.20 [etienne@localhost linux-2.6.20]$ cp ../../linux2/linux-2.6.20-gujin/.config . [etienne@localhost linux-2.6.20]$ make oldconfig > log [etienne@localhost linux-2.6.20]$ make vmlinux >> log ... few warnings ... [etienne@localhost linux-2.6.20]$ readelf -l vmlinux Elf file type is EXEC (Executable file) Entry point 0x100000 There are 3 program headers, starting at offset 52 Program Headers: Type Offset VirtAddr PhysAddr FileSiz MemSiz Flg Align LOAD 0x001000 0xc0100000 0x00100000 0x3995e0 0x3995e0 R E 0x1000 LOAD 0x39b000 0xc049a000 0x0049a000 0x75085 0xaf000 RWE 0x1000 NOTE 0x000000 0x00000000 0x00000000 0x00000 0x00000 R 0x4 Section to Segment mapping: Segment Sections... 00 .text __ex_table .rodata .pci_fixup __param 01 .data .data_nosave .data.page_aligned .data.cacheline_aligned .data.read_mostly .data.init_task .init.text .init.data .init.setup .initcall.init .con_initcall.init .security_initcall.init .altinstructions .altinstr_replacement .exit.text .init.ramfs .bss 02 [etienne@localhost linux-2.6.20]$ OK, everything is fine. The ld script affects manually the program header by: /* arch/i386/kernel/vmlinux.lds.S */ PHDRS { text PT_LOAD FLAGS(5); /* R_E */ data PT_LOAD FLAGS(7); /* RWE */ note PT_NOTE FLAGS(4); /* R__ */ } And the text section goes to the :text program header, it is linked at a fixed address. .text : AT(ADDR(.text) - LOAD_OFFSET) { ... } :text = 0x9090 Now I want to create another text section - I was thinking that a new program header would be added or an error message because I create it a another address, using the patch at sourceforge at: http://sourceforge.net/project/showfiles.php?group_id=15465&package_id=174813&release_id=482035 The important extract is: SECTIONS { .realmode 0 : AT (ADDR(.bss) + SIZEOF(.bss) - LOAD_OFFSET) { ... all my sections ... } :text = 0x9090 . = 0xc0000000 ; /* or more */ .text : AT(ADDR(.text) - LOAD_OFFSET) { ... original part ... } :text = 0x9090 [etienne@localhost linux-2.6.20]$ patch -p1 < ../../linux/gujin-patch-2.6.20-rc5 patching file arch/i386/Kconfig patching file arch/i386/kernel/Makefile patching file arch/i386/kernel/realmode.c patching file arch/i386/kernel/setup.c patching file arch/i386/kernel/vmlinux.lds.S patching file arch/i386/Makefile patching file include/asm-i386/realmode.h patching file Makefile patching file scripts/gzcopy.c patching file scripts/gzparam.c [etienne@localhost linux-2.6.20]$ make linux-2.6.20.kgz > log But nothing change and I do not get any warning or error for a section containing code not linked to a program header: [etienne@localhost linux-2.6.20]$ readelf -l vmlinux Elf file type is EXEC (Executable file) Entry point 0x100000 There are 3 program headers, starting at offset 52 Program Headers: Type Offset VirtAddr PhysAddr FileSiz MemSiz Flg Align LOAD 0x001000 0xc0100000 0x00100000 0x44a830 0x44a830 R E 0x1000 LOAD 0x44c000 0xc049a000 0x0049a000 0x75085 0xaf000 RWE 0x1000 NOTE 0x000000 0x00000000 0x00000000 0x00000 0x00000 R 0x4 Section to Segment mapping: Segment Sections... 00 .bss .text __ex_table .rodata .pci_fixup __param .data .data_nosave .data.page_aligned .data.cacheline_aligned .data.read_mostly .data.init_task .init.text .init.data .init.setup .initcall.init .con_initcall.init .security_initcall.init .altinstructions .altinstr_replacement .exit.text .init.ramfs 01 .bss .data .data_nosave .data.page_aligned .data.cacheline_aligned .data.read_mostly .data.init_task .init.text .init.data .init.setup .initcall.init .con_initcall.init .security_initcall.init .altinstructions .altinstr_replacement .exit.text .init.ramfs 02 [etienne@localhost linux-2.6.20]$ And there is something wrong once using strip from this file: [etienne@localhost linux-2.6.20]$ objcopy -S vmlinux vmlinux.stripped BFD: vmlinux.stripped: warning: allocated section `.realmode' not in segment [etienne@localhost linux-2.6.20]$ readelf -l vmlinux.stripped Elf file type is EXEC (Executable file) Entry point 0x100000 There are 3 program headers, starting at offset 52 Program Headers: Type Offset VirtAddr PhysAddr FileSiz MemSiz Flg Align LOAD 0x001000 0xc0100000 0x00100000 0x40f085 0x449000 R E 0x1000 LOAD 0x411000 0xc049a000 0x0049a000 0x75085 0xaf000 RWE 0x1000 NOTE 0x000000 0x00000000 0x00000000 0x00000 0x00000 R 0x4 Section to Segment mapping: Segment Sections... 00 .bss .text __ex_table .rodata .pci_fixup __param .data .data_nosave .data.page_aligned .data.cacheline_aligned .data.read_mostly .data.init_task .init.text .init.data .init.setup .initcall.init .con_initcall.init .security_initcall.init .altinstructions .altinstr_replacement .exit.text .init.ramfs 01 .bss .data .data_nosave .data.page_aligned .data.cacheline_aligned .data.read_mostly .data.init_task .init.text .init.data .init.setup .initcall.init .con_initcall.init .security_initcall.init .altinstructions .altinstr_replacement .exit.text .init.ramfs 02 [etienne@localhost linux-2.6.20]$ Note that the first program header MemSiz overlap the second program header: 0x00100000 + 0x449000 > 0x0049a000 And the two FileSiz are not aligned so the memset at PhysAddr+FileSiz is done at an unaligned address so will be slow. I do not know what should be the behaviour, just something seems wrong. Something completely different, when using the linux patch I posted on LKM, where a new program header is created for this section, and is removed if i want to get rid of the section .realmode completely because I want to generate the old target, you can ask me the question: [etienne@localhost linux-2.6.20-gujin]$ objcopy -R .note -R .comment -R .realmode -S vmlinux vmlinux.tmp BFD: vmlinux: warning: Empty loadable segment detected, is this intentional ? [etienne@localhost linux-2.6.20-gujin]$ My answer is that it is not needed to keep this empty loadable segment, you can strip it. I do not care of the warning because that command is used "-O binary" by Linux Makefile and does not produce the warning. Thanks, Etienne.
It is very hard for people to reproduce it exactly. Please provide all necessary input object files. I will take a look.
I have the vmlinux ELF file here, but it is bigger than the maximum attachement size. Is there a public ftp server useable or shall I mail it somewhere? (compressed: 2.5 Mbytes).
The problem is the linker script: .realmode 0 : AT (ADDR(.bss) + SIZEOF(.bss) - 0xC0000000) { ... } :text = 0x9090 You put .realmode in the text segment. But its address doesn't belong to the text segment. I suggest you put .realmode in a separate segment.
I have already put .realmode in a separate segment, that was not the problem. I just would have liked a linker warning/error in this case, instead of droping silently the section, and having an "unsupported configuration" i.e. this vmlinuz file which cannot "strip".
Linker should issue an error when a section is placed into a segment improperly: [hjl@gnu-2 objcopy-2]$ cat foo.t OUTPUT_FORMAT("elf32-i386", "elf32-i386", "elf32-i386") OUTPUT_ARCH(i386) PHDRS { text PT_LOAD FLAGS(5); /* R_E */ data PT_LOAD FLAGS(7); /* RWE */ } SECTIONS { . = 0xC0000000 + ((0x100000 + (0x100000 - 1)) & ~(0x100000 - 1)); .bar : AT(ADDR(.bar) - 0xC0000000) { *(.bar) } :text .bss : AT(ADDR(.bss) - 0xC0000000) { *(.bss) } .foo 0 : AT(ADDR(.bss) + SIZEOF(.bss) - 0xC0000000) { *(.foo) } :text /DISCARD/ : { *(.*) } } [hjl@gnu-2 objcopy-2]$ cat foo.s .section .bar,"ax","progbits" .byte 0 .section .foo,"aw","progbits" .byte 0 .bss .long 0 [hjl@gnu-2 objcopy-2]$ make copy as --32 -o foo.o foo.s ./ld -m elf_i386 -o foo -T foo.t foo.o objcopy foo foo-cpy BFD: foo-cpy: warning: allocated section `.foo' not in segment [hjl@gnu-2 objcopy-2]$ make readelf -lS foo There are 7 section headers, starting at offset 0x1034: Section Headers: [Nr] Name Type Addr Off Size ES Flg Lk Inf Al [ 0] NULL 00000000 000000 000000 00 0 0 0 [ 1] .bar PROGBITS c0100000 001000 000001 00 AX 0 0 1 [ 2] .bss NOBITS c0100004 001001 000004 00 WA 0 0 4 [ 3] .foo PROGBITS 00000000 001008 000001 00 WA 0 0 1 [ 4] .shstrtab STRTAB 00000000 001009 00002a 00 0 0 1 [ 5] .symtab SYMTAB 00000000 00114c 000040 10 6 4 4 [ 6] .strtab STRTAB 00000000 00118c 000001 00 0 0 1 Key to Flags: W (write), A (alloc), X (execute), M (merge), S (strings) I (info), L (link order), G (group), x (unknown) O (extra OS processing required) o (OS specific), p (processor specific) Elf file type is EXEC (Executable file) Entry point 0x0 There are 1 program headers, starting at offset 52 Program Headers: Type Offset VirtAddr PhysAddr FileSiz MemSiz Flg Align LOAD 0x001000 0xc0100000 0x00100000 0x00009 0x00010 R E 0x1000 Section to Segment mapping: Segment Sections... 00 .bar .bss [hjl@gnu-2 objcopy-2]$
A patch is posted at http://sourceware.org/ml/binutils/2007-02/msg00375.html
Patch works nicely (did it on 2.17 vanilia): /home/etienne/linux/binutils-2.17/ld/ld-new: vmlinux: section `.realmode' can't be allocated in segement 0 /home/etienne/linux/binutils-2.17/ld/ld-new: final link failed: Bad value Just maybe "s/segement/segment/", but do not take me as a reference in English. Thanks.
Fixed by http://sourceware.org/ml/binutils/2007-03/msg00308.html