Linux/i386 boot protocol version 2.03
The following patch is a fairly small and fully backwards compatible change to the i386 boot protocol. It makes the maximum legal initrd address explicitly available to the boot loader, so it doesn't have to guess. To make matters worse, the current documentation specifies 0x3C000000 as the top address (exclusive), but the real address is 0x38000000.
This patch:
a) Bumps the boot protocol version number to 2.03;
b) Adds a field to the boot header which contains the maximum legal initrd address;
c) Slightly reorganizes a couple of macros to make (b) possible;
d) Documents this change and the actual behaviour for previous protocol versions.
The patch is against 2.4.17-pre6, but applies to 2.5.1-pre8 as well.
-hpa
Documentation/i386/boot.txt renamed from linux-2.4.17-pre6/Documentation/i386/boot.txt +23 -12
@@ -2,7 +2,7 @@ ---------------------------- H. Peter Anvin <[email protected]> - Last update 2000-10-29 + Last update 2001-12-09 On the i386 platform, the Linux kernel uses a rather complicated boot convention. This has evolved partially due to historical aspects, as
@@ -25,12 +25,15 @@ Protocol 2.01: (Kernel 1.3.76) Added a heap overrun warning. Protocol 2.02: (Kernel 2.4.0-test3-pre3) New command line protocol. - Lower the conventional memory ceiling. No overwrite + Lower the conventional memory ceiling. No overwrite of the traditional setup area, thus making booting safe for systems which use the EBDA from SMM or 32-bit BIOS entry points. zImage deprecated but still supported. +Protocol 2.03: (???) Explicitly makes the highest possible initrd address + available to the bootloader. + **** MEMORY LAYOUT
@@ -45,7 +48,7 @@ 098000 +------------------------+ | Kernel setup | The kernel real-mode code. 090200 +------------------------+ - | Kernel boot sector | The kernel legacy boot sector. + | Kernel boot sector | The kernel legacy boot sector. 090000 +------------------------+ | Protected-mode kernel | The bulk of the kernel image. 010000 +------------------------+
@@ -62,7 +65,7 @@ When using bzImage, the protected-mode kernel was relocated to 0x100000 ("high memory"), and the kernel real-mode block (boot sector, setup, and stack/heap) was made relocatable to any address between -0x10000 and end of low memory. Unfortunately, in protocols 2.00 and +0x10000 and end of low memory. Unfortunately, in protocols 2.00 and 2.01 the command line is still required to live in the 0x9XXXX memory range, and that memory range is still overwritten by the early kernel. The 2.02 protocol fixes that.
@@ -71,7 +74,7 @@ low memory touched by the boot loader -- as low as possible, since some newer BIOSes have begun to allocate some rather large amounts of memory, called the Extended BIOS Data Area, near the top of low -memory. The boot loader should use the "INT 12h" BIOS call to verify +memory. The boot loader should use the "INT 12h" BIOS call to verify how much low memory is available. Unfortunately, if INT 12h reports that the amount of memory is too
@@ -123,6 +126,7 @@ 0224/2 2.01+ heap_end_ptr Free memory after setup end 0226/2 N/A pad1 Unused 0228/4 2.02+ cmd_line_ptr 32-bit pointer to the kernel command line +022C/4 2.03+ initrd_addr_max Highest legal initrd address For backwards compatibility, if the setup_sects field contains 0, the real value is 4.
@@ -180,9 +184,9 @@ The initrd should typically be located as high in memory as possible, as it may otherwise get overwritten by the early - kernel initialization sequence. However, it must never be - located above address 0x3C000000 if you want all kernels to - read it. + kernel initialization sequence. However, it must never be + located above the address specified in the initrd_addr_max + field. The initrd should be at least 4K page aligned. cmd_line_ptr: If the protocol version is 2.02 or higher, this is a 32-bit
@@ -192,7 +196,15 @@ command line, in which case you can point this to an empty string (or better yet, to the string "auto".) If this field is left at zero, the kernel will assume that your boot loader - does not support the 2.02 protocol. + does not support the 2.02+ protocol. + + ramdisk_max: + The maximum address that may be occupied by the initrd + contents. For boot protocols 2.02 or earlier, this field is + not present, and the maximum address is 0x37FFFFFF. (This + address is defined as the address of the highest safe byte, so + if your ramdisk is exactly 131072 bytes long and this field is + 0x37FFFFFF, you can start your ramdisk at 0x37FE0000.) **** THE KERNEL COMMAND LINE
@@ -254,14 +266,14 @@ if ( protocol >= 0x0202 ) { cmd_line_ptr = base_ptr + 0x9000; } else { - cmd_line_magic = 0xA33F; + cmd_line_magic = 0xA33F; cmd_line_offset = 0x9000; setup_move_size = 0x9100; } } else { /* Very old kernel */ - cmd_line_magic = 0xA33F; + cmd_line_magic = 0xA33F; cmd_line_offset = 0x9000; /* A very old kernel MUST have its real-mode code
@@ -411,4 +423,3 @@ After completing your hook, you should jump to the address that was in this field before your boot loader overwrote it. -
arch/i386/boot/setup.S renamed from linux-2.4.17-pre6/arch/i386/boot/setup.S +8 -3
@@ -50,7 +50,8 @@ #include <linux/compile.h> #include <asm/boot.h> #include <asm/e820.h> - +#include <asm/page.h> + /* Signature words to ensure LILO loaded us right */ #define SIG1 0xAA55 #define SIG2 0x5A5A
@@ -79,7 +80,7 @@ # This is the setup header, and it must start at %cs:2 (old 0x9020:2) .ascii "HdrS" # header signature - .word 0x0202 # header version number (>= 0x0105) + .word 0x0203 # header version number (>= 0x0105) # or else old loadlin-1.5 will fail) realmode_swtch: .word 0, 0 # default_switch, SETUPSEG start_sys_seg: .word SYSSEG
@@ -153,6 +154,10 @@ # can be located anywhere in # low memory 0x10000 or higher. +ramdisk_max: .long __MAXMEM-1 # (Header version 0x0203 or later) + # The highest safe address for + # the contents of an initrd + trampoline: call start_of_setup .space 1024 # End of setup header #####################################################
@@ -539,7 +544,7 @@ cmpw $0, %cs:realmode_swtch jz rmodeswtch_normal - lcall *%cs:realmode_swtch + lcall %cs:realmode_swtch jmp rmodeswtch_end
arch/i386/kernel/setup.c renamed from linux-2.4.17-pre6/arch/i386/kernel/setup.c +1 -3
@@ -827,10 +827,8 @@#define PFN_PHYS (x) ((x) << PAGE_SHIFT)/* -* 128MB for vmalloc and initrd +* Reserved space for vmalloc and iomap - defined in asm/page.h */ -#define VMALLOC_RESERVE (unsigned long)(128 << 20) -#define MAXMEM (unsigned long)(-PAGE_OFFSET-VMALLOC_RESERVE)#define MAXMEM_PFN PFN_DOWN(MAXMEM)#define MAX_NONPAE_PFN (1 << 20)
include/asm-i386/page.h renamed from linux-2.4.17-pre6/include/asm-i386/page.h +9 -0
@@ -80,6 +80,12 @@#define __PAGE_OFFSET (0xC0000000) +/* +* This much address space is reserved for vmalloc() and iomap() +* as well as fixmap mappings. +*/ +#define __VMALLOC_RESERVE (128 << 20) +#ifndef __ASSEMBLY__/*
@@ -118,6 +124,9 @@#endif /* __ASSEMBLY__ */ #define PAGE_OFFSET ((unsigned long)__PAGE_OFFSET) +#define VMALLOC_RESERVE ((unsigned long)__VMALLOC_RESERVE) +#define __MAXMEM (-__PAGE_OFFSET-__VMALLOC_RESERVE) +#define MAXMEM ((unsigned long)(-PAGE_OFFSET-VMALLOC_RESERVE))#define __pa (x) ((unsigned long)(x)-PAGE_OFFSET)#define __va (x) ((void *)((unsigned long)(x)+PAGE_OFFSET))#define virt_to_page (kaddr) (mem_map + (__pa(kaddr) >> PAGE_SHIFT))
Signature · 3 lines
<[email protected]> at work, <[email protected]> in private! "Unix gives you enough rope to shoot yourself in the foot." http://www.zytor.com/~hpa/puzzle.txt
Quoted 7 lines · H. Peter Anvin
The following patch is a fairly small and fully backwards compatible change to the i386 boot protocol. It makes the maximum legal initrd address explicitly available to the boot loader, so it doesn't have to guess. To make matters worse, the current documentation specifies
0x3C000000as the top address (exclusive), but the real address is0x38000000.
Quoted 8 lines
This patch:
a) Bumps the boot protocol version number to 2.03;
b) Adds a field to the boot header which contains the maximum legal initrd address;
c) Slightly reorganizes a couple of macros to make (b) possible;
d) Documents this change and the actual behaviour for previous protocol versions.
This looks reasonable.
A couple of notes:
1) The minimum safe ramdisk address is 8MB (since 2.4.10). On low mem machines you can get away with placing a ramdisk lower. But we don't do any checking in our initial 8MB memory map.
2) If we use units of kilobytes instead of bytes for this we don't loose any precision and gain the ability to put a ramdisk in high memory without bumping the protocol version.
3) If we are going to export the maximum address we should also export the minimum address.
Eric
Quoted 11 lines · Eric W. Biederman
A couple of notes:
1) The minimum safe ramdisk address is 8MB (since 2.4.10). On low mem machines you can get away with placing a ramdisk lower. But we don't do any checking in our initial 8MB memory map.
2) If we use units of kilobytes instead of bytes for this we don't loose any precision and gain the ability to put a ramdisk in high memory without bumping the protocol version.
3) If we are going to export the maximum address we should also export the minimum address.
(2) I guess I'm not so concerned with the ramdisk in highmem since it is extrememly unlikely any boot loader will be able to take advantage of that. It could be an issue for x86-64, I guess.
(3) Contradicts (1) as well as issues with older kernels. Keep in mind what happens if you violate this limit: the bootloader should be loading the initrd as high as possible, so the only difference is if you get the error message from the boot loader or from the kernel later. If you're going to export a limit, you better make sure it's right; "8MB except on low memory configurations" doesn't cut it. It's exactly on those low memory configurations that this limit matters *at all*.
-hpa
Quoted 3 lines · Eric W. Biederman
2) If we use units of kilobytes instead of bytes for this we don't loose any precision and gain the ability to put a ramdisk in high memory without bumping the protocol version.
Thought about it some more, and then realized we would have to do a lot more overhaul of the boot protocol than this to support initrd in highmem ... and we'd still not be able to actually use it on any real configuration. Thanks, but no thanks.
-hpa
Quoted 12 lines · H. Peter Anvin
Quoted 10 lines · Eric W. Biederman
A couple of notes:
1) The minimum safe ramdisk address is 8MB (since 2.4.10). On low mem machines you can get away with placing a ramdisk lower. But we don't do any checking in our initial 8MB memory map.
2) If we use units of kilobytes instead of bytes for this we don't loose any precision and gain the ability to put a ramdisk in high memory without bumping the protocol version.
3) If we are going to export the maximum address we should also export the minimum address.
Quoted 3 lines
(2) I guess I'm not so concerned with the ramdisk in highmem since it is extrememly unlikely any boot loader will be able to take advantage of that. It could be an issue for x86-64, I guess.
I agree. When you can spend a little extra time and get something working. I find it nice. If it doesn't work out oh, well.
Quoted 3 lines
(3) Contradicts (1) as well as issues with older kernels. Keep in mind what happens if you violate this limit: the bootloader should be loading the initrd as high as possible
Hmm. Bootloaders have had lots of restrictions on getting up high. Plus in some real sense it make sense to pack the two as close together as possible. For 2.5.x and Al Viro's initfs work I'd like to do that.
If you load your ramdisk at the same location every time, after a point you have fewer surprises, and potentially a simpler bootloader.
Quoted 7 lines
, so the only difference is if you get the error message from
the boot loader or from the kernel later. If you're going to export a limit, you better make sure it's right; "8MB except on low memory configurations" doesn't cut it. It's exactly on those low memory configurations that this limit
matters *at all*.
8MB is the safe limit. The only worry some case right now is the bootmem allocator with a darn huge bitmap. It is allocated dynamically, and it is extremely hard to predict where that will be except in the initial page tables 8MB in size. If the ugly bitmap is allocated elsewhere the kernel can't use it.
On low memory configurations you have no choice. But it will probably work. Getting better requires more work in setup.c than is likely to go into 2.4.x.
On old kernels the big worry is the allocation of the struct page array. I have observed this array exceed 64MB in memory. High enough so that some older bootloaders can't load initrd high enough to cope. I believe int 0x15 function ah=87 can't move blocks above that.
Either going high or going low with ramdisk placement requires knowledge of how the kernel allocates it's memory. And people don't often reference bootloader docs when they change the policy so it is fragile to depend upon how the kernel does it's allocations.
So if we are going to fix the fragility please let's handle both ends. Then it will be o.k. to put the ramdisk where ever the kernel says it is safe. And we can stop this guessing and breaking game.
Eric
Quoted 6 lines · Eric W. Biederman
Quoted 5 lines
(3) Contradicts (1) as well as issues with older kernels. Keep in mind what happens if you violate this limit: the bootloader should be loading the initrd as high as possible
Quoted 7 lines
Hmm. Bootloaders have had lots of restrictions on getting up high. Plus in some real sense it make sense to pack the two as close together as possible. For 2.5.x and Al Viro's initfs work I'd like
to do that.If you load your ramdisk at the same location every time, after a point you have fewer surprises, and potentially a simpler bootloader.
NO, PLEASE PLEASE PLEASE DON'T DO THAT!!!!
Not only would it royally fuck over people with small amounts of RAM, it would also really fuck over people who use the Linux boot protocol for other purposes, which is getting quite common.
Quoted 7 lines
Quoted 7 lines
, so the only difference is if you get the error message from
the boot loader or from the kernel later. If you're going to export a limit, you better make sure it's right; "8MB except on low memory configurations" doesn't cut it. It's exactly on those low memory configurations that this limit
Quoted 9 lines
Quoted 2 lines
matters *at all*.
8MB is the safe limit. The only worry some case right now is the bootmem allocator with a darn huge bitmap. It is allocated dynamically, and it is extremely hard to predict where that will be except in the initial page tables 8MB in size. If the ugly bitmap is allocated elsewhere the kernel can't use it.
"Safe limit" isn't applicable here. It either works or it doesn't. You're breaking systems which it would otherwise work on, which is not acceptable.
Quoted 3 lines
So if we are going to fix the fragility please let's handle both ends. Then it will be o.k. to put the ramdisk where ever the kernel says it is safe. And we can stop this guessing and breaking game.
You're giving the boot loader options that are really inappropriate. All that will do is result in more variance between boot loaders and the resulting bugs that will bite some bootloaders and not others.
Allowing unneeded options in protocols is a source of bugs. You seem to think this is a good idea, it's not.
-hpa
Quoted 2 lines · H. Peter Anvin
Allowing unneeded options in protocols is a source of bugs. You seem to think this is a good idea, it's not.
I think it is more that we disagree on what are unneeded options, and
what kinds of bugs happen. Additionally there are issues with
which things are mandatory, and which things are recommended behavior, in our communications.
I think we agree on goals of getting the booting as simple as error free as possible.
Beyond that it looks like our world views are so different we do not successfully communicate, so unless I have code or a specific case that needs fixing, I will not try.
Eric