Re: [LinuxBIOS] Complete and generic 32bit/64bit support

[Jordan Crouse]

On 21/11/07 13:37 -0800, yhlu wrote:
> On Nov 21, 2007 12:03 PM, Myles Watson  wrote:
> > > On Wed, Nov 21, 2007 at 12:13:23PM -0700, Myles Watson wrote:
> > > > It makes the same changes as mentioned before, but it also uses the
> > > > correct kernel image depending on the architecture.  mkelfimage
> > > > doesn't work with an x86_64 vmlinux, and with an i386 bzImage it
> > > > generates a _much_ larger elf.
>
> for x86_64 support
> there is a patch i posted several months ago.

Excellent!  This should fit into buildrom very nicely.
Thanks.

Jordan
> [PATCH] mkelfImage: Take elf64 vmlinux and produce elf32
>
> change to use e_entry instead of hardcode for vmlinux
> if e_entry is not equal to paddr of fist phdr, will use paddr instead, and not switch to 64bit in convert
> otherwise will use e_entry and switch to 64bit in convert
>
> also fix the bss_size calculation.
>
> Signed-off-By: Yinghai Lu
>
> diff -uNr mkelfImage-2.7/linux-i386/convert.h mkelfImage-2.7.new/linux-i386/convert.h
> --- mkelfImage-2.7/linux-i386/convert.h	2006-03-17 06:08:21.000000000 -0800
> +++ mkelfImage-2.7.new/linux-i386/convert.h	2006-11-08 12:32:15.000000000 -0800
> @@ -3,6 +3,8 @@
>  #define CONVERTLOC 		0x10000
>  #define REAL_MODE_DATA_LOC	0x20000
>  #define GDTLOC	   		0x21000
> +#define GDT64LOC		0x22000
> +#define PGTLOC			0x23000
>
> #define DEFAULT_ROOT_DEV ((0x3<<8)| 0)
>
> @@ -18,10 +20,13 @@
>  struct image_parameters {
>    uint32_t convert_magic; /* a signature to verify mkelfImage was built properly */
>    uint32_t gdt_size;
> uint32_t gdt64_size;
> uint32_t pgt_size;
> uint32_t bss_size;
> uint16_t ramdisk_flags;
> uint16_t root_dev;
> uint32_t entry;
> uint32_t switch_64;
> uint32_t initrd_start;
> uint32_t initrd_size;
> uint8_t  cmdline[CMDLINE_MAX];
>
> diff -uNr mkelfImage-2.7/linux-i386/convert.lds mkelfImage-2.7.new/linux-i386/convert.lds
> --- mkelfImage-2.7/linux-i386/convert.lds	2003-01-16 13:41:14.000000000 -0800
> +++ mkelfImage-2.7.new/linux-i386/convert.lds	2006-11-08 12:19:12.000000000 -0800
> @@ -27,7 +27,7 @@
>    	*(.bss)
>    }
>    _end = . ;
> bss_size = _end - _bss;
>
> bss_sizex = _end - _bss;
>
> /DISCARD/ : {
> *(.comment)
>
> diff -uNr mkelfImage-2.7/linux-i386/head.S mkelfImage-2.7.new/linux-i386/head.S
> --- mkelfImage-2.7/linux-i386/head.S	2003-01-24 14:48:03.000000000 -0800
> +++ mkelfImage-2.7.new/linux-i386/head.S	2006-11-08 12:19:32.000000000 -0800
> @@ -59,7 +59,7 @@
>    movl	$GDTLOC, %edi
>    movl	$(gdt_end - gdt), %ecx
>    rep	movsb
> # Linux makes stupid assumptions about the segments
>  # that are already setup, so setup a new gdt & ldt
>  # and then reload the segment registers.
>
> @@ -95,22 +95,113 @@
>    # Flag to indicate we are the bootstrap processor
>    xorl	%ebx, %ebx
> movl    switch_64, %eax
> cmp	$1, %eax
> jz	switch_to_64
> # Clear the unspecified registers for good measure
>  xorl	%eax, %eax
>  xorl	%ecx, %ecx
>  xorl	%edx, %edx
>  xorl	%edi, %edi
> xorl	%esp, %esp
> xorl	%ebp, %ebp
> # do not clear esp, we still need to use lret later
>    pushl $PROT_CODE_SEG
>    movl entry, %eax
>    pushl %eax
> lret
>
> +switch_to_64:
> /* We need to switch to 64bit before use startup_64 entry go to kernel */
> /*
> Prepare for entering 64 bit mode
> */
>    # Move the gdt64 where Linux will not smash it during decompression
> movl	%esi, %eax # save the real mode pointer
> movl    $gdt64, %esi
> movl    $GDT64LOC, %edi
> movl    $(gdt64_end - gdt64), %ecx
> rep     movsb
> movl	%eax, %esi
> /* Load new GDT with the 64bit segments using 32bit descriptor */
> lgdt	gdt64
> /* Enable PAE mode */
> xorl    %eax, %eax
> btsl    $5, %eax
> movl    %eax, %cr4
> /*
> Build early 4G boot pagetable
> */
>   /* Initialize Page tables to 0*/
>   movl    $PGTLOC, %edi
>   xorl    %eax, %eax
>   movl    $((4096*6)/4), %ecx
>   rep     stosl
>   /* Build Level 4 */
>   movl    $(PGTLOC + 0), %edi
>   leal    0x1007 (%edi), %eax
>   movl    %eax, 0(%edi)
>   /* Build Level 3 */
>   movl    $(PGTLOC + 0x1000), %edi
>   leal    0x1007(%edi), %eax
>   movl    $4, %ecx
>
> +1:     movl    %eax, 0x00(%edi)
>   addl    $0x00001000, %eax
>   addl    $8, %edi
>   decl    %ecx
>   jnz     1b
>   /* Build Level 2 */
>   movl    $(PGTLOC + 0x2000), %edi
>   movl    $0x00000183, %eax
>   movl    $2048, %ecx
>
> +1:     movl    %eax, 0(%edi)
>   addl    $0x00200000, %eax
>   addl    $8, %edi
>   decl    %ecx
>   jnz     1b
>   /* Enable the boot page tables */
>   movl    $PGTLOC, %eax
>   movl    %eax, %cr3
>   /* Enable Long mode in EFER (Extended Feature Enable Register) */
>   movl    $0xc0000080, %ecx
>   rdmsr
>   btsl    $8, %eax
>   wrmsr
> /* Preparing for 64bit jmp */
>    pushl $PROT_CODE_SEG
>    movl entry, %eax
>    pushl %eax
>   /* Enter paged protected Mode, activating Long Mode */
>    xorl    %eax, %eax
>    btsl    $31, %eax
>    btsl    $0, %eax
>    movl    %eax, %cr0
>    /*
>     * At this point we're in long mode but in 32bit compatibility mode
>     * with EFER.LME = 1, CS.L = 0, CS.D = 1 (and in turn
>     * EFER.LMA = 1). Now we want to jump in 64bit mode, to do that we use
>     * the new gdt/idt that has __KERNEL_CS with CS.L = 1.
>     */
>
> # Jump to the linux kernel
> ljmp	$ PROT_CODE_SEG , $ 0x100000
>
> lret
>
> /* Routines to query the BIOS... */
>
> /**************************************************************************
>  E820_MEMSIZE - Get a listing of memory regions
>  **************************************************************************/
> @@ -409,6 +500,7 @@
>    .word	0x9200				# data read/write
>    .word	0x00CF				# granularity = 4096, 386
>    					#  (+5th nibble of limit)
>
> /*
> The layout of the per-CPU GDT under Linux:
>
> @@ -446,7 +538,14 @@
>   */
>  gdt_end:
>
> +gdt64:
> .word	gdt64_end - gdt64
> .long	GDT64LOC
> .word	0
> .quad	0x0000000000000000	/* NULL descriptor */
> .quad	0x00af9a000000ffff	/* __KERNEL_CS */
> .quad	0x00cf92000000ffff	/* __KERNEL_DS */
>
> +gdt64_end:
>
> .section ".trailer", "a"
>    /* Constants set at build time, these are at the very end of my image */
> @@ -458,14 +557,20 @@
>    .long	CONVERT_MAGIC
>  gdt_size:
>    .long	gdt_end - gdt
> +gdt64_size:
> .long	gdt64_end - gdt64
>
> +pgt_size:
> .long   4096*6
>
> bss_size:
> .long	bss_size
> .long	bss_sizex
>
> ramdisk_flags:
>    .word	0
>  root_dev:
>    .word	DEFAULT_ROOT_DEV
>  entry:
>    .long	0
> +switch_64:
> .long	0
>
> initrd_start:
>    .long	0
>  initrd_size:
> diff -uNr mkelfImage-2.7/linux-i386/mkelf-linux-i386.c mkelfImage-2.7.new/linux-i386/mkelf-linux-i386.c
> --- mkelfImage-2.7/linux-i386/mkelf-linux-i386.c	2006-03-17 06:08:22.000000000 -0800
> +++ mkelfImage-2.7.new/linux-i386/mkelf-linux-i386.c	2006-11-08 09:54:58.000000000 -0800
> @@ -19,13 +19,16 @@
>  static void (*parse_kernel_type)(struct kernel_info *info, char *kernel_buf, size_t kernel_size);
>  static void parse_bzImage_kernel(struct kernel_info *info, char *kernel_buf, size_t kernel_size);
>  static void parse_elf32_kernel(struct kernel_info *info, char *kernel_buf, size_t kernel_size);
> +static void parse_elf64_kernel(struct kernel_info *info, char *kernel_buf, size_t kernel_size);
>
> +char *vmlinux_x86_64_probe(char *kernel_buf, off_t kernel_size);
>
> char *vmlinux_i386_probe(char *kernel_buf, off_t kernel_size)
>  {
>    Elf32_Ehdr *ehdr;
>    Elf32_Phdr *phdr;
>    int i;
> int hdr1, hdr2;
> int phdrs;
> ehdr = (Elf32_Ehdr *)kernel_buf;
> if (
> (ehdr->e_ident[EI_MAG0] != ELFMAG0) ||
>
> @@ -35,7 +38,8 @@
>    	return "No ELF signature found on kernel\n";
>    }
>    if (ehdr->e_ident[EI_CLASS] != ELFCLASS32) {
> return "Not a 32bit ELF kernel\n";
> return vmlinux_x86_64_probe(kernel_buf, kernel_size);
>
> +//		return "Not a 32bit ELF kernel\n";
>    }
>    if (ehdr->e_ident[EI_DATA]  != ELFDATA2LSB) {
>    	return "Not a little endian ELF kernel\n";
> @@ -54,27 +58,64 @@
>    	return "Kernel uses bad program header size.\n";
>    }
>    phdr = (Elf32_Phdr *)(kernel_buf + le32_to_cpu(ehdr->e_phoff));
> hdr1 = hdr2 = -1;
> for(i = 0; i < le32_to_cpu(ehdr->e_phnum); i++) {
> phdrs = 0;
> for(i = 0; i < le16_to_cpu(ehdr->e_phnum); i++) {
> if (le32_to_cpu(phdr[i].p_type) != PT_LOAD)
> 	continue;
> if (((hdr1 != -1) && 
> 	((le32_to_cpu(phdr[hdr1].p_paddr) & 0xfffffff) != 0x100000)) ||
> 	(hdr2 != -1)) {
> 	return "Too many PT_LOAD segments to be a linux kernel\n";
> }
> if (hdr1 == -1) {
> 	hdr1 = i;
> } else {
> 	hdr2 = i;
> }
> phdrs++;
> }
> if (hdr1 == -1) {
> if (phdrs == 0) {
> return "No PT_LOAD segments!\n";
> }
> parse_kernel_type = parse_elf32_kernel;
> return 0;
>
> }
> +char *vmlinux_x86_64_probe(char *kernel_buf, off_t kernel_size)
> +{
>    Elf64_Ehdr *ehdr;
>    Elf64_Phdr *phdr;
>    int i;
> int phdrs = 0;
>    ehdr = (Elf64_Ehdr *)kernel_buf;
>    if (
>            (ehdr->e_ident[EI_MAG0] != ELFMAG0) ||
>            (ehdr->e_ident[EI_MAG1] != ELFMAG1) ||
>            (ehdr->e_ident[EI_MAG2] != ELFMAG2) ||
>            (ehdr->e_ident[EI_MAG3] != ELFMAG3)) {
>            return "No ELF signature found on kernel\n";
>    }
>    if (ehdr->e_ident[EI_CLASS] != ELFCLASS64) {
>            return "Not a 64bit ELF kernel\n";
>    }
>    if (ehdr->e_ident[EI_DATA]  != ELFDATA2LSB) {
>            return "Not a little endian ELF kernel\n";
>    }
>    if (le16_to_cpu(ehdr->e_type) != ET_EXEC) {
>            return "Not an executable kernel\n";
>    }
>    if (le16_to_cpu(ehdr->e_machine) != EM_X86_64) {
>            return "Not an x86_64 kernel\n";
>    }
>    if (    (ehdr->e_ident[EI_VERSION] != EV_CURRENT) ||
>            (le32_to_cpu(ehdr->e_version) != EV_CURRENT)) {
>            return "Kernel not using ELF version 1.\n";
>    }
>    if (le16_to_cpu(ehdr->e_phentsize) != sizeof(*phdr)) {
>            return "Kernel uses bad program header size.\n";
>    }
>    phdr = (Elf64_Phdr *)(kernel_buf + le64_to_cpu(ehdr->e_phoff));
> phdrs = 0;
>    for(i = 0; i < le16_to_cpu(ehdr->e_phnum); i++) {
>            if (le32_to_cpu(phdr[i].p_type) != PT_LOAD)
>                    continue;
> phdrs++;
>    }
>    if (phdrs == 0) {
>            return "No PT_LOAD segments!\n";
>    }
>    parse_kernel_type = parse_elf64_kernel;
>    return 0;
>
> +}
>
> char *bzImage_i386_probe(char *kernel_buf, off_t kernel_size)
>  {
> @@ -118,17 +159,14 @@
>
> struct kernel_info
>  {
> void *kernel;
> size_t filesz;
> size_t memsz;
> size_t paddr;
> size_t vaddr;
> void *kernel2;
> size_t filesz2;
> size_t memsz2;
> size_t paddr2;
> size_t vaddr2;
> int phdrs;
> void *kernel[4];
> size_t filesz[4];
> size_t memsz[4];
> size_t paddr[4];
> size_t vaddr[4];
> size_t entry;
> size_t switch_64;
> char *version;
>
> };
>
> @@ -137,43 +175,68 @@
>    Elf32_Ehdr *ehdr;
>    Elf32_Phdr *phdr;
>    int i;
> int hdr1, hdr2;
> int phdrs;
> ehdr = (Elf32_Ehdr *)kernel_buf;
> phdr = (Elf32_Phdr *)(kernel_buf + ehdr->e_phoff);
> hdr1 = hdr2 = -1;
> phdrs = 0;
> for(i = 0; i < le16_to_cpu(ehdr->e_phnum); i++) {
> if (le32_to_cpu(phdr[i].p_type) != PT_LOAD)
> 	continue;
> if (hdr2 != -1) {
> 	die("Too many PT_LOAD segments to be a linux kernel\n");
> }
> if (hdr1 == -1) {
> 	hdr1 = i;
> } else {
> 	hdr2 = i;
> }
> }
> if (hdr1 == -1) {
> die("No PT_LOAD segments!\n");
> info->kernel[phdrs]  = kernel_buf + le32_to_cpu(phdr[i].p_offset);
> info->filesz[phdrs]  = le32_to_cpu(phdr[i].p_filesz);
> info->memsz[phdrs]   = le32_to_cpu(phdr[i].p_memsz);
> info->paddr[phdrs]   = le32_to_cpu(phdr[i].p_paddr) & 0xfffffff;
> info->vaddr[phdrs]   = le32_to_cpu(phdr[i].p_vaddr);
> phdrs++;
> }
> info->kernel  = kernel_buf + le32_to_cpu(phdr[hdr1].p_offset);
> info->filesz  = le32_to_cpu(phdr[hdr1].p_filesz);
> info->memsz   = le32_to_cpu(phdr[hdr1].p_memsz);
> info->paddr   = le32_to_cpu(phdr[hdr1].p_paddr) & 0xfffffff;
> info->vaddr   = le32_to_cpu(phdr[hdr1].p_vaddr);
> if (hdr2 != -1) {
> info->kernel2 = kernel_buf + le32_to_cpu(phdr[hdr2].p_offset);
> info->filesz2 = le32_to_cpu(phdr[hdr2].p_filesz);
> info->memsz2  = le32_to_cpu(phdr[hdr2].p_memsz);
> info->paddr2  = le32_to_cpu(phdr[hdr2].p_paddr) & 0xfffffff;
> info->vaddr2  = le32_to_cpu(phdr[hdr2].p_vaddr);
> if(!phdrs)
> die("We need at least one phdr\n");
> info->phdrs = phdrs;
>    info->entry   = le32_to_cpu(ehdr->e_entry);
> info->switch_64   = 0; //not convert from elf64
> info->version = "unknown";
>
> +}
>
> +static void parse_elf64_kernel(struct kernel_info *info, char *kernel_buf, size_t kernel_size)
> +{
>    Elf64_Ehdr *ehdr;
>    Elf64_Phdr *phdr;
>    int i;
> int phdrs;
>    ehdr = (Elf64_Ehdr *)kernel_buf;
>    phdr = (Elf64_Phdr *)(kernel_buf + le64_to_cpu(ehdr->e_phoff));
> phdrs = 0;
> for(i = 0; i < le16_to_cpu(ehdr->e_phnum); i++) {
> if (le32_to_cpu(phdr[i].p_type) != PT_LOAD)
> 	continue;
> info->kernel[phdrs]  = kernel_buf + le64_to_cpu(phdr[i].p_offset);
> info->filesz[phdrs]  = le64_to_cpu(phdr[i].p_filesz);
> info->memsz[phdrs]   = le64_to_cpu(phdr[i].p_memsz);
> info->paddr[phdrs]   = le64_to_cpu(phdr[i].p_paddr) & 0xffffff;
> info->vaddr[phdrs]   = le64_to_cpu(phdr[i].p_vaddr);
> phdrs++;
> }
> if(!phdrs)
> die("We need at least one phdr\n");
>
> info->entry   = 0x100000;
> info->version = "unknown";
> info->phdrs = phdrs;
>    info->entry   = le64_to_cpu(ehdr->e_entry);
>
> +#if  0
> if (info->entry != info->paddr[0]) {
> info->entry = info->paddr[0]; // we still have startup_32 there
> info->switch_64   = 0; //not convert from elf64
> } else
>
> +#endif
> info->switch_64   = 1; //convert from elf64
>    info->version = "unknown";
>
> }
>
> static void parse_bzImage_kernel(struct kernel_info *info, char *kernel_buf, size_t kernel_size)
>  {
>    struct x86_linux_header *hdr;
> @@ -186,12 +249,14 @@
>    }
>    offset = 512 + (512 *setup_sects);
> info->kernel  = kernel_buf + offset;
> info->filesz  = kernel_size - offset;
> info->memsz   = 0x700000;
> info->paddr   = 0x100000;
> info->vaddr   = 0x100000;
> info->entry   = info->paddr;
> info->kernel[0]  = kernel_buf + offset;
> info->filesz[0]  = kernel_size - offset;
> info->memsz[0]   = 0x700000;
> info->paddr[0]   = 0x100000;
> info->vaddr[0]   = 0x100000;
> info->phdrs = 1;
> info->entry   = info->paddr[0];
> info->switch_64   = 0; //not convert from elf64, even later bzImage become elf64, it still includes startup_32
> info->version = kernel_buf + 512 + le16_to_cpu(hdr->kver_addr);
>
> }
>
> @@ -237,6 +302,7 @@
>    struct kernel_info kinfo;
>    struct image_parameters *params;
>    int index;
>
> int i;
>
> int opt;
> static const struct option options[] = {
>
> @@ -310,7 +376,7 @@
>
> /* Add a program header for the note section */
>    index = 4;
> index += kinfo.kernel2 ? 1:0;
> index += (kinfo.phdrs - 1);
> index += ramdisk_size ? 1:0;
> phdr = add_program_headers(ehdr, index);
>
> @@ -328,23 +394,24 @@
>    phdr[2].p_paddr  = REAL_MODE_DATA_LOC;
>    phdr[2].p_vaddr  = REAL_MODE_DATA_LOC;
>    phdr[2].p_filesz = 0;
> phdr[2].p_memsz  = (GDTLOC - REAL_MODE_DATA_LOC) + params->gdt_size;
> if(!kinfo.switch_64)
> phdr[2].p_memsz = (GDTLOC - REAL_MODE_DATA_LOC) + params->gdt_size;
> else
> phdr[2].p_memsz = (PGTLOC - REAL_MODE_DATA_LOC) + params->pgt_size;
> phdr[2].p_data   = 0;
> phdr[3].p_paddr  = kinfo.paddr;
> phdr[3].p_vaddr  = kinfo.vaddr;
> phdr[3].p_filesz = kinfo.filesz;
> phdr[3].p_memsz  = kinfo.memsz;
> phdr[3].p_data   = kinfo.kernel;
> if( (phdr[1].p_paddr + phdr[1].p_memsz) > phdr[2].p_paddr) {
> die("Internal error: need to increase REAL_MODE_DATA_LOC !\n");
> }
> index = 4;
> index = 3;
> /* Put the second kernel frament if present */
> if (kinfo.kernel2) {
> phdr[index].p_paddr  = kinfo.paddr2;
> phdr[index].p_vaddr  = kinfo.vaddr2;
> phdr[index].p_filesz = kinfo.filesz2;
> phdr[index].p_memsz  = kinfo.memsz2;
> phdr[index].p_data   = kinfo.kernel2;
> for(i=0;i<kinfo.phdrs;i++) {
> phdr[index].p_paddr  = kinfo.paddr[i];
> phdr[index].p_vaddr  = kinfo.vaddr[i];
> phdr[index].p_filesz = kinfo.filesz[i];
> phdr[index].p_memsz  = kinfo.memsz[i];
> phdr[index].p_data   = kinfo.kernel[i];
> index++;
> }
>
> @@ -352,6 +419,10 @@
>     */
>    params->initrd_start = params->initrd_size = 0;
>    if (ramdisk_size) {
> if( (phdr[index-1].p_paddr + phdr[index-1].p_memsz) > ramdisk_base) {
> 	die("need to increase increase ramdisk_base !\n");
> }
> phdr[index].p_paddr  = ramdisk_base;
>   phdr[index].p_vaddr  = ramdisk_base;
>   phdr[index].p_filesz = ramdisk_size;
>
> @@ -364,6 +435,7 @@
>
> /* Set the start location */
>    params->entry = kinfo.entry;
>
> params->switch_64 = kinfo.switch_64;
> ehdr->e_entry = phdr[1].p_paddr;
>
> /* Setup the elf notes */
>
> diff -uNr mkelfImage-2.7/linux-ia64/mkelf-linux-ia64.c mkelfImage-2.7.new/linux-ia64/mkelf-linux-ia64.c
> --- mkelfImage-2.7/linux-ia64/mkelf-linux-ia64.c	2006-03-17 10:09:16.000000000 -0800
> +++ mkelfImage-2.7.new/linux-ia64/mkelf-linux-ia64.c	2006-11-07 18:14:27.000000000 -0800
> @@ -81,7 +81,7 @@
>    phdr = (Elf64_Phdr *)(kernel_buf + le64_to_cpu(ehdr->e_phoff));
>    phdrs = 0;
>    for(i = 0; i < le16_to_cpu(ehdr->e_phnum); i++) {
> if (le16_to_cpu(phdr[i].p_type) != PT_LOAD)
> if (le32_to_cpu(phdr[i].p_type) != PT_LOAD)
> continue;
> phdrs++;
> }
>