[PATCH v2] x86, boot: Allow 64bit EFI kernel to be loaded above 4G

[PATCH v2] x86, boot: Allow 64bit EFI kernel to be loaded above 4G

[Yinghai Lu]

Now could use kexec to place kernel/boot_params/cmd_line/initrd
above 4G, but that is with legacy interface with startup_64 directly.

This patch will allow 64bit EFI kernel to be loaded above 4G
and use EFI HANDOVER PROTOCOL to start the kernel.

Current 32bit code32_start is used for passing around load address,
so it will overflow when kernel is loaded abover 4G.

The patch mainly add ext_code32_start to take load address high 32bits.

After this patch, could use patched grub2-x86_64.efi to place
kernel/boot_params/cmd_line/initrd all above 4G and execute the kernel
above 4G.

bootlog like:

kernel: done                           [ linux  9.25MiB  100%  6.66MiB/s ]
params: [1618fc000,1618fffff]
cmdline: [1618fb000,1618fb7fe]
kernel: [15e000000,161385fff]
initrd: [15bcbe000,15dffffbb]
initrd: 1 file done             [ initrd.img  35.26MiB  100%  11.93MiB/s ]
early console in decompress_kernel
decompress_kernel:
  input: [0x15fd0b3b4-0x16063c803], output: 0x15e000000, heap: [0x160645b00-0x16064daff]

Decompressing Linux... xz... Parsing ELF... done.
Booting the kernel.
[    0.000000] bootconsole [uart0] enabled
[    0.000000]    real_mode_data :      phys 00000001618fc000
[    0.000000]    real_mode_data :      virt ffff8801618fc000
[    0.000000] Kernel Layout:
[    0.000000]   .text: [0x15e000000-0x15f08f72c]
[    0.000000] .rodata: [0x15f200000-0x15fa44fff]
[    0.000000]   .data: [0x15fc00000-0x15fe545ff]
[    0.000000]   .init: [0x15fe56000-0x16021afff]
[    0.000000]    .bss: [0x160229000-0x16135ffff]
[    0.000000]    .brk: [0x161360000-0x161385fff]
[    0.000000] memblock_reserve: [0x0000000009f000-0x000000000fffff] flags 0x0 * BIOS reserved
...
[    0.000000] memblock_reserve: [0x0000015e000000-0x0000016135ffff] flags 0x0 TEXT DATA BSS
[    0.000000] memblock_reserve: [0x0000015bcbe000-0x0000015dffffff] flags 0x0 RAMDISK

-v2: add cast to avoid warning with 32bit, also update description for
     ext_code32_start in boot.txt

Signed-off-by: Yinghai Lu <yinghai@kernel.org>

---
 Documentation/x86/boot.txt            |   19 +++++++++++++++++++
 arch/x86/boot/compressed/eboot.c      |   15 ++++++++++-----
 arch/x86/boot/compressed/head_64.S    |    7 ++++++-
 arch/x86/boot/header.S                |    3 ++-
 arch/x86/include/uapi/asm/bootparam.h |    1 +
 arch/x86/kernel/asm-offsets.c         |    1 +
 6 files changed, 39 insertions(+), 7 deletions(-)

Index: linux-2.6/arch/x86/include/uapi/asm/bootparam.h
===================================================================
--- linux-2.6.orig/arch/x86/include/uapi/asm/bootparam.h
+++ linux-2.6/arch/x86/include/uapi/asm/bootparam.h
@@ -84,6 +84,7 @@ struct setup_header {
 	__u64	pref_address;
 	__u32	init_size;
 	__u32	handover_offset;
+	__u32	ext_code32_start;
 } __attribute__((packed));
 
 struct sys_desc_table {
Index: linux-2.6/arch/x86/kernel/asm-offsets.c
===================================================================
--- linux-2.6.orig/arch/x86/kernel/asm-offsets.c
+++ linux-2.6/arch/x86/kernel/asm-offsets.c
@@ -68,6 +68,7 @@ void common(void) {
 	OFFSET(BP_kernel_alignment, boot_params, hdr.kernel_alignment);
 	OFFSET(BP_pref_address, boot_params, hdr.pref_address);
 	OFFSET(BP_code32_start, boot_params, hdr.code32_start);
+	OFFSET(BP_ext_code32_start, boot_params, hdr.ext_code32_start);
 
 	BLANK();
 	DEFINE(PTREGS_SIZE, sizeof(struct pt_regs));
Index: linux-2.6/arch/x86/boot/compressed/head_64.S
===================================================================
--- linux-2.6.orig/arch/x86/boot/compressed/head_64.S
+++ linux-2.6/arch/x86/boot/compressed/head_64.S
@@ -264,6 +264,8 @@ ENTRY(efi_pe_entry)
 	mov	%rax, %rsi
 	leaq	startup_32(%rip), %rax
 	movl	%eax, BP_code32_start(%rsi)
+	shr	$32, %rax
+	movl	%eax, BP_ext_code32_start(%rsi)
 	jmp	2f		/* Skip the relocation */
 
 handover_entry:
@@ -287,7 +289,10 @@ fail:
 	hlt
 	jmp	fail
 2:
-	movl	BP_code32_start(%esi), %eax
+	movl	BP_code32_start(%rsi), %eax
+	movl	BP_ext_code32_start(%rsi), %ebx
+	shl	$32, %rbx
+	orq	%rbx, %rax
 	leaq	preferred_addr(%rax), %rax
 	jmp	*%rax
 
Index: linux-2.6/arch/x86/boot/compressed/eboot.c
===================================================================
--- linux-2.6.orig/arch/x86/boot/compressed/eboot.c
+++ linux-2.6/arch/x86/boot/compressed/eboot.c
@@ -1388,6 +1388,7 @@ struct boot_params *efi_main(struct efi_
 	void *handle;
 	efi_system_table_t *_table;
 	bool is64;
+	unsigned long loaded_addr;
 
 	efi_early = c;
 
@@ -1429,9 +1430,12 @@ struct boot_params *efi_main(struct efi_
 	 * If the kernel isn't already loaded at the preferred load
 	 * address, relocate it.
 	 */
-	if (hdr->pref_address != hdr->code32_start) {
-		unsigned long bzimage_addr = hdr->code32_start;
-		status = efi_relocate_kernel(sys_table, &bzimage_addr,
+	loaded_addr = hdr->code32_start;
+	loaded_addr |= (unsigned long)((u64)hdr->ext_code32_start << 32);
+	if (hdr->pref_address != loaded_addr) {
+		unsigned long loaded_addr_orig = loaded_addr;
+
+		status = efi_relocate_kernel(sys_table, &loaded_addr,
 					     hdr->init_size, hdr->init_size,
 					     hdr->pref_address,
 					     hdr->kernel_alignment);
@@ -1440,8 +1444,9 @@ struct boot_params *efi_main(struct efi_
 			goto fail;
 		}
 
-		hdr->pref_address = hdr->code32_start;
-		hdr->code32_start = bzimage_addr;
+		hdr->pref_address = loaded_addr_orig;
+		hdr->code32_start = loaded_addr & 0xffffffff;
+		hdr->ext_code32_start = (unsigned long)((u64)loaded_addr >> 32);
 	}
 
 	status = exit_boot(boot_params, handle, is64);
Index: linux-2.6/arch/x86/boot/header.S
===================================================================
--- linux-2.6.orig/arch/x86/boot/header.S
+++ linux-2.6/arch/x86/boot/header.S
@@ -301,7 +301,7 @@ _start:
 	# Part 2 of the header, from the old setup.S
 
 		.ascii	"HdrS"		# header signature
-		.word	0x020d		# header version number (>= 0x0105)
+		.word	0x020e		# header version number (>= 0x0105)
 					# or else old loadlin-1.5 will fail)
 		.globl realmode_swtch
 realmode_swtch:	.word	0, 0		# default_switch, SETUPSEG
@@ -449,6 +449,7 @@ pref_address:		.quad LOAD_PHYSICAL_ADDR
 #endif
 init_size:		.long INIT_SIZE		# kernel initialization size
 handover_offset:	.long 0			# Filled in by build.c
+ext_code32_start:	.long 0			# werid one!
 
 # End of setup header #####################################################
 
Index: linux-2.6/Documentation/x86/boot.txt
===================================================================
--- linux-2.6.orig/Documentation/x86/boot.txt
+++ linux-2.6/Documentation/x86/boot.txt
@@ -61,6 +61,9 @@ Protocol 2.12:	(Kernel 3.8) Added the xl
 	 	to struct boot_params for loading bzImage and ramdisk
 		above 4G in 64bit.
 
+Protocol 2.14:	(Kernel 3.20) Added the ext_code32_start to support 64bit
+		EFI kernel to be loaded above 4G.
+
 **** MEMORY LAYOUT
 
 The traditional memory map for the kernel loader, used for Image or
@@ -197,6 +200,7 @@ Offset	Proto	Name		Meaning
 0258/8	2.10+	pref_address	Preferred loading address
 0260/4	2.10+	init_size	Linear memory required during initialization
 0264/4	2.11+	handover_offset	Offset of handover entry point
+0268/4	2.14+	ext_code32_start	Extended part for code32_start
 
 (1) For backwards compatibility, if the setup_sects field contains 0, the
     real value is 4.
@@ -738,6 +742,14 @@ Offset/size:	0x264/4
 
   See EFI HANDOVER PROTOCOL below for more details.
 
+Field name:	ext_code32_start
+Type:		modify (optional, reloc)
+Offset/size:	0x268/4
+Protocol:	2.14+
+
+  This field is the upper 32bits of load address when EFI 64bit kernel
+  is loaded above 4G. And it is used with code32_start to compare to
+  pref_address to decide if kernel need to be relocated further.
 
 **** THE IMAGE CHECKSUM
 
@@ -1122,4 +1134,11 @@ The boot loader *must* fill out the foll
     o hdr.ramdisk_image (if applicable)
     o hdr.ramdisk_size  (if applicable)
 
+for 64bit, when loading above 4G, *must* fill out the following fields,
+
+    o hdr.ext_code32_start
+    o ext_cmd_line_ptr
+    o ext_ramdisk_image (if applicable)
+    o ext_ramdisk_size  (if applicable)
+
 All other fields should be zero.

Re: [PATCH v2] x86, boot: Allow 64bit EFI kernel to be loaded above 4G

[Matt Fleming]

On Sun, 22 Feb, at 07:43:48PM, Yinghai Lu wrote:
> Index: linux-2.6/arch/x86/boot/header.S
> ===================================================================
> --- linux-2.6.orig/arch/x86/boot/header.S
> +++ linux-2.6/arch/x86/boot/header.S
> @@ -301,7 +301,7 @@ _start:
>  	# Part 2 of the header, from the old setup.S
>  
>  		.ascii	"HdrS"		# header signature
> -		.word	0x020d		# header version number (>= 0x0105)
> +		.word	0x020e		# header version number (>= 0x0105)
>  					# or else old loadlin-1.5 will fail)
>  		.globl realmode_swtch
>  realmode_swtch:	.word	0, 0		# default_switch, SETUPSEG
> @@ -449,6 +449,7 @@ pref_address:		.quad LOAD_PHYSICAL_ADDR
>  #endif
>  init_size:		.long INIT_SIZE		# kernel initialization size
>  handover_offset:	.long 0			# Filled in by build.c
> +ext_code32_start:	.long 0			# werid one!
  
How about a comment like, "# Upper 32-bits of code32_start" ?

> Index: linux-2.6/Documentation/x86/boot.txt
> ===================================================================
> --- linux-2.6.orig/Documentation/x86/boot.txt
> +++ linux-2.6/Documentation/x86/boot.txt
> @@ -61,6 +61,9 @@ Protocol 2.12:	(Kernel 3.8) Added the xl
>  	 	to struct boot_params for loading bzImage and ramdisk
>  		above 4G in 64bit.
>  
> +Protocol 2.14:	(Kernel 3.20) Added the ext_code32_start to support 64bit
> +		EFI kernel to be loaded above 4G.
> +
>  **** MEMORY LAYOUT
>  
>  The traditional memory map for the kernel loader, used for Image or
> @@ -197,6 +200,7 @@ Offset	Proto	Name		Meaning
>  0258/8	2.10+	pref_address	Preferred loading address
>  0260/4	2.10+	init_size	Linear memory required during initialization
>  0264/4	2.11+	handover_offset	Offset of handover entry point
> +0268/4	2.14+	ext_code32_start	Extended part for code32_start
>  
>  (1) For backwards compatibility, if the setup_sects field contains 0, the
>      real value is 4.
> @@ -738,6 +742,14 @@ Offset/size:	0x264/4
>  
>    See EFI HANDOVER PROTOCOL below for more details.
>  
> +Field name:	ext_code32_start
> +Type:		modify (optional, reloc)
> +Offset/size:	0x268/4
> +Protocol:	2.14+
> +
> +  This field is the upper 32bits of load address when EFI 64bit kernel
> +  is loaded above 4G. And it is used with code32_start to compare to
> +  pref_address to decide if kernel need to be relocated further.

I think we can delete the second sentence.

But these are minimal changes and I'm happy to fix them up when applying
this patch if that's OK?

-- 
Matt Fleming, Intel Open Source Technology Center

Re: [PATCH v2] x86, boot: Allow 64bit EFI kernel to be loaded above 4G

[Yinghai Lu]

On Tue, Feb 24, 2015 at 1:55 PM, Matt Fleming <matt@codeblueprint.co.uk> wrote:
> On Sun, 22 Feb, at 07:43:48PM, Yinghai Lu wrote:
>> +Field name:  ext_code32_start
>> +Type:                modify (optional, reloc)
>> +Offset/size: 0x268/4
>> +Protocol:    2.14+
>> +
>> +  This field is the upper 32bits of load address when EFI 64bit kernel
>> +  is loaded above 4G. And it is used with code32_start to compare to
>> +  pref_address to decide if kernel need to be relocated further.
>
> I think we can delete the second sentence.
>
> But these are minimal changes and I'm happy to fix them up when applying
> this patch if that's OK?

Sure. Please go ahead.

Thanks

Yinghai

Re: [PATCH v2] x86, boot: Allow 64bit EFI kernel to be loaded above 4G

[Matt Fleming]

On Tue, 24 Feb, at 04:31:27PM, Yinghai Lu wrote:
> On Tue, Feb 24, 2015 at 1:55 PM, Matt Fleming <matt-mF/unelCI9GS6iBeEJttW/XRex20P6io@public.gmane.org> wrote:
> > On Sun, 22 Feb, at 07:43:48PM, Yinghai Lu wrote:
> >> +Field name:  ext_code32_start
> >> +Type:                modify (optional, reloc)
> >> +Offset/size: 0x268/4
> >> +Protocol:    2.14+
> >> +
> >> +  This field is the upper 32bits of load address when EFI 64bit kernel
> >> +  is loaded above 4G. And it is used with code32_start to compare to
> >> +  pref_address to decide if kernel need to be relocated further.
> >
> > I think we can delete the second sentence.
> >
> > But these are minimal changes and I'm happy to fix them up when applying
> > this patch if that's OK?
> 
> Sure. Please go ahead.

Applied, thanks Yinghai.

-- 
Matt Fleming, Intel Open Source Technology Center

Re: [PATCH v2] x86, boot: Allow 64bit EFI kernel to be loaded above 4G

[Ingo Molnar]

* Matt Fleming <matt@codeblueprint.co.uk> wrote:

> On Sun, 22 Feb, at 07:43:48PM, Yinghai Lu wrote:
> > Index: linux-2.6/arch/x86/boot/header.S
> > ===================================================================
> > --- linux-2.6.orig/arch/x86/boot/header.S
> > +++ linux-2.6/arch/x86/boot/header.S
> > @@ -301,7 +301,7 @@ _start:
> >  	# Part 2 of the header, from the old setup.S
> >  
> >  		.ascii	"HdrS"		# header signature
> > -		.word	0x020d		# header version number (>= 0x0105)
> > +		.word	0x020e		# header version number (>= 0x0105)
> >  					# or else old loadlin-1.5 will fail)
> >  		.globl realmode_swtch
> >  realmode_swtch:	.word	0, 0		# default_switch, SETUPSEG
> > @@ -449,6 +449,7 @@ pref_address:		.quad LOAD_PHYSICAL_ADDR
> >  #endif
> >  init_size:		.long INIT_SIZE		# kernel initialization size
> >  handover_offset:	.long 0			# Filled in by build.c
> > +ext_code32_start:	.long 0			# werid one!
>   
> How about a comment like, "# Upper 32-bits of code32_start" ?

Also, in the Linux kernel we should not go about inventing 
new English words like 'werid', right?

Thanks,

	Ingo

Re: [PATCH v2] x86, boot: Allow 64bit EFI kernel to be loaded above 4G

[Matt Fleming]

On Wed, 25 Feb, at 11:31:32AM, Ingo Molnar wrote:
> 
> * Matt Fleming <matt-mF/unelCI9GS6iBeEJttW/XRex20P6io@public.gmane.org> wrote:
> 
> > On Sun, 22 Feb, at 07:43:48PM, Yinghai Lu wrote:
> > > Index: linux-2.6/arch/x86/boot/header.S
> > > ===================================================================
> > > --- linux-2.6.orig/arch/x86/boot/header.S
> > > +++ linux-2.6/arch/x86/boot/header.S
> > > @@ -301,7 +301,7 @@ _start:
> > >  	# Part 2 of the header, from the old setup.S
> > >  
> > >  		.ascii	"HdrS"		# header signature
> > > -		.word	0x020d		# header version number (>= 0x0105)
> > > +		.word	0x020e		# header version number (>= 0x0105)
> > >  					# or else old loadlin-1.5 will fail)
> > >  		.globl realmode_swtch
> > >  realmode_swtch:	.word	0, 0		# default_switch, SETUPSEG
> > > @@ -449,6 +449,7 @@ pref_address:		.quad LOAD_PHYSICAL_ADDR
> > >  #endif
> > >  init_size:		.long INIT_SIZE		# kernel initialization size
> > >  handover_offset:	.long 0			# Filled in by build.c
> > > +ext_code32_start:	.long 0			# werid one!
> >   
> > How about a comment like, "# Upper 32-bits of code32_start" ?
> 
> Also, in the Linux kernel we should not go about inventing 
> new English words like 'werid', right?

Yeah, my intention is to replace "werid one" with the comment I
suggested above.

-- 
Matt Fleming, Intel Open Source Technology Center