Yinghai Lu <yinghai at kernel.org> writes:
> On Sat, Nov 17, 2012 at 10:53 PM, Yinghai Lu <yinghai at kernel.org> wrote:
>> On Sat, Nov 17, 2012 at 10:50 PM, Eric W. Biederman
>> <ebiederm at xmission.com> wrote:
>>>>
>>>> so we should two purgatory for x86_64 ?
>>>> like one for 64bit and one for 32bit/16bit
>>>
>>> The problem is that the current 32bit assembly code is not possition
>>> independent. If we can rewrite that assembly code to be position
>>> independent no relocs will be generated and we should be good.
>>>
>>> The 16bit entry point code was already written as position independent
>>> code so it should not cause problems.
>>>
>>> Just having one piece of code to deal with (if we can figure it out)
>>> looks to be the simpler and more maintainable solution.
>>
>> sure.
>
> just resent -v2 patches for kernel parts.
>
> attached are changes for kexec-tools without put 64 bit purgartory above 4g...
>
> hope you can work out 32bit position independent.
It looks like this is enough to fix purgatory.
I have tested this in so far as that it doesn't generate relocs, but I
haven't had a test to see if the code actually works.
YH can you verify this boots both 32bit and 64bit kernels for you?
Thanks,
Eric
diff --git a/purgatory/Makefile b/purgatory/Makefile
index ee1679c..e39adec 100644
--- a/purgatory/Makefile
+++ b/purgatory/Makefile
@@ -64,6 +64,7 @@ $(PURGATORY): $(PURGATORY_OBJS)
$(CC) $(CFLAGS) $(LDFLAGS) -o $@ $^
# $(LD) $(LDFLAGS) $(EXTRA_LDFLAGS) --no-undefined -e purgatory_start -r -o $@ $(PURGATORY_OBJS) $(UTIL_LIB)
+ $(STRIP) --strip-debug $@
echo::
@echo "PURGATORY_SRCS $(PURGATORY_SRCS)"
diff --git a/purgatory/arch/i386/entry32-16-debug.S b/purgatory/arch/i386/entry32-16-debug.S
index 82b58ca..2f47607 100644
--- a/purgatory/arch/i386/entry32-16-debug.S
+++ b/purgatory/arch/i386/entry32-16-debug.S
@@ -29,15 +29,17 @@
.balign 16
entry16_debug:
.code32
- /* Compute where I am running at */
- movl $entry16_debug, %ebx
+ /* Compute where I am running at (assumes esp valid) */
+ call 1f
+1: popl %ebx
/* Fixup my real mode segment */
movl %ebx, %eax
shrl $4, %eax
- movw %ax, 2 + realptr
+ movw %ax, (2 + realptr - entry16_debug)(%ebx)
/* Fixup the gdt */
+ movl %ebx, (0x02 + gdt - entry16_debug)(%ebx)
movl %ebx, %eax
shll $16, %eax
@@ -49,19 +51,19 @@ entry16_debug:
andl $0xff000000, %edx
orl %edx, %ecx
- orl %eax, 0x08 + gdt
- orl %ecx, 0x0c + gdt
- orl %eax, 0x10 + gdt
- orl %ecx, 0x14 + gdt
+ orl %eax, (0x08 + gdt - entry16_debug)(%ebx)
+ orl %ecx, (0x0c + gdt - entry16_debug)(%ebx)
+ orl %eax, (0x10 + gdt - entry16_debug)(%ebx)
+ orl %ecx, (0x14 + gdt - entry16_debug)(%ebx)
DEBUG_CHAR('a')
/* Setup the classic BIOS interrupt table at 0x0 */
- lidt idtptr
+ lidt (idtptr - entry16_debug)(%ebx)
DEBUG_CHAR('b')
/* Provide us with 16bit segments that we can use */
- lgdt gdt
+ lgdt (gdt - entry16_debug)(%ebx)
DEBUG_CHAR('c')
/* Note we don't disable the a20 line, (this shouldn't be required)
@@ -160,7 +162,7 @@ idtptr:
gdt:
/* 0x00 unusable segment so used as the gdt ptr */
.word gdt_end - gdt - 1
- .long gdt
+ .long 0 /* gdt */
.word 0
/* 0x08 16 bit real mode code segment */
diff --git a/purgatory/arch/i386/entry32-16.S b/purgatory/arch/i386/entry32-16.S
index aaf1273..20a1ce6 100644
--- a/purgatory/arch/i386/entry32-16.S
+++ b/purgatory/arch/i386/entry32-16.S
@@ -24,15 +24,17 @@
.balign 16
entry16:
.code32
- /* Compute where I am running at */
- movl $entry16, %ebx
+ /* Compute where I am running at (assumes esp valid) */
+ call 1f
+1: popl %ebx
/* Fixup my real mode segment */
movl %ebx, %eax
shrl $4, %eax
- movw %ax, 2 + realptr
+ movw %ax, (2 + realptr - entry16)(%ebx)
/* Fixup the gdt */
+ movl %ebx, (0x02 + gdt - entry16)(%ebx)
movl %ebx, %eax
shll $16, %eax
@@ -44,17 +46,17 @@ entry16:
andl $0xff000000, %edx
orl %edx, %ecx
- orl %eax, 0x08 + gdt
- orl %ecx, 0x0c + gdt
- orl %eax, 0x10 + gdt
- orl %ecx, 0x14 + gdt
+ orl %eax, (0x08 + gdt - entry16)(%ebx)
+ orl %ecx, (0x0c + gdt - entry16)(%ebx)
+ orl %eax, (0x10 + gdt - entry16)(%ebx)
+ orl %ecx, (0x14 + gdt - entry16)(%ebx)
/* Setup the classic BIOS interrupt table at 0x0 */
- lidt idtptr
+ lidt (idtptr - entry16)(%ebx)
/* Provide us with 16bit segments that we can use */
- lgdt gdt
+ lgdt (gdt - entry16)(%ebx)
/* Note we don't disable the a20 line, (this shouldn't be required)
* The code to do it is in kexec_test and it is a real pain.
@@ -147,7 +149,7 @@ idtptr:
gdt:
/* 0x00 unusable segment so used as the gdt ptr */
.word gdt_end - gdt - 1
- .long gdt
+ .long 0 /* gdt */
.word 0
/* 0x08 16 bit real mode code segment */
diff --git a/purgatory/arch/x86_64/Makefile b/purgatory/arch/x86_64/Makefile
index 22b4228..7300937 100644
--- a/purgatory/arch/x86_64/Makefile
+++ b/purgatory/arch/x86_64/Makefile
@@ -16,9 +16,11 @@ dist += purgatory/arch/x86_64/Makefile $(x86_64_PURGATORY_SRCS_native) \
purgatory/arch/x86_64/purgatory-x86_64.h
# Don't add sources in i386/ to dist, as i386/Makefile adds them
-x86_64_PURGATORY_SRCS += purgatory/arch/i386/entry32-16.S
+x86_64_PURGATORY_SRCS += purgatory/arch/i386/entry32-16.S
x86_64_PURGATORY_SRCS += purgatory/arch/i386/entry32-16-debug.S
x86_64_PURGATORY_SRCS += purgatory/arch/i386/crashdump_backup.c
x86_64_PURGATORY_SRCS += purgatory/arch/i386/console-x86.c
x86_64_PURGATORY_SRCS += purgatory/arch/i386/vga.c
x86_64_PURGATORY_SRCS += purgatory/arch/i386/pic.c
+
+x86_64_PURGATORY_EXTRA_CFLAGS = -mcmodel=large
diff --git a/purgatory/arch/x86_64/entry64-32.S b/purgatory/arch/x86_64/entry64-32.S
index 66f8a85..0d394ad 100644
--- a/purgatory/arch/x86_64/entry64-32.S
+++ b/purgatory/arch/x86_64/entry64-32.S
@@ -24,13 +24,34 @@
.equ CR0_PG, 0x80000000
.text
+ .balign 16
.globl entry32, entry32_regs
entry32:
.code64
- /* Setup a gdt that should that is generally usefully */
+ /* Setup the 4G offset of entry32 lm_exit code segment */
+ movq $0x00CF9A000000ffff, %rax
+
+ leaq entry32(%rip), %rbx /* Low 24 bits */
+ andq $0xffffff, %rbx
+ shlq $16, %rbx
+ orq %rbx, %rax
+
+ leaq entry32(%rip), %rbx /* High 8 bits */
+ movq $0xff000000, %rdx
+ andq %rdx, %rbx
+ shlq $32, %rbx
+ orq %rbx, %rax
+
+ movq %rax, (gdt + 0x20)(%rip)
+
+ /* Setup a gdt that is generally usefully */
lgdt gdt(%rip)
-
+
+ /* Setup the far pointer to the entry point */
+ movl eip(%rip), %eax
+ movl %eax, entry32_addr(%rip)
+
/* Switch to 32bit compatiblity mode */
ljmp *lm_exit_addr(%rip)
lm_exit:
@@ -60,19 +81,19 @@ lm_exit:
movl %eax, %gs
/* Load the registers */
- movl eax, %eax
- movl ecx, %ecx
- movl edx, %edx
- movl esi, %esi
- movl edi, %edi
- movl esp, %esp
- movl ebp, %ebp
- movl ebx, %ebx
+ movl %cs:eax - entry32, %eax
+ movl %cs:ecx - entry32, %ecx
+ movl %cs:edx - entry32, %edx
+ movl %cs:esi - entry32, %esi
+ movl %cs:edi - entry32, %edi
+ movl %cs:esp - entry32, %esp
+ movl %cs:ebp - entry32, %ebp
+ movl %cs:ebx - entry32, %ebx
/* Jump to the loaded image */
- jmpl *(eip)
+ jmpl *%cs:entry32_addr - entry32
- .section ".rodata"
+ .section ".data"
.balign 16
gdt: /* 0x00 unusable segment
* 0x08 unused
@@ -88,8 +109,8 @@ gdt: /* 0x00 unusable segment
/* 0x18 4GB flat data segment */
.word 0xFFFF, 0x0000, 0x9200, 0x00CF
- /* 0x20 dummy */
- .word 0x0000, 0x0000, 0x0000, 0x000
+ /* 0x20 4GB flat code segment base at entry32 */
+ .word 0xFFFF, 0x0000, 0x9A00, 0x0CF
/* 0x28 dummy */
.word 0x0000, 0x0000, 0x0000, 0x000
/* 0x30 dummy */
@@ -115,9 +136,15 @@ gdt_end:
.section ".rodata"
.balign 4
lm_exit_addr:
- .long lm_exit
- .long 0x10
-
+ .long lm_exit - entry32
+ .long 0x20
+
+ .section ".data"
+ .balign 4
+entry32_addr:
+ .long 0x00000000
+ .long 0x10
+
.section ".rodata"
.balign 4
entry32_regs:
@@ -129,6 +156,9 @@ esi: .long 0x00000000
edi: .long 0x00000000
esp: .long 0x00000000
ebp: .long 0x00000000
-eip: .long entry16
- .size entry32_regs, . - entry32_regs
+eip: .quad entry16 /* low 32 bits address
+ * high 32bits zeros
+ * uses 64bit reloc
+ */
+ .size entry32_regs, (. - 4) - entry32_regs
diff --git a/purgatory/arch/x86_64/entry64.S b/purgatory/arch/x86_64/entry64.S
index 666023c..e3223b7 100644
--- a/purgatory/arch/x86_64/entry64.S
+++ b/purgatory/arch/x86_64/entry64.S
@@ -37,9 +37,10 @@ entry64:
movl %eax, %fs
movl %eax, %gs
- movq $stack_init, %rsp
+ leaq stack_init(%rip), %rsp
pushq $0x10 /* CS */
- pushq $new_cs_exit
+ leaq new_cs_exit(%rip), %rax
+ pushq %rax
lretq
new_cs_exit:
diff --git a/purgatory/arch/x86_64/setup-x86_64.S b/purgatory/arch/x86_64/setup-x86_64.S
index 74997fa..95572d8 100644
--- a/purgatory/arch/x86_64/setup-x86_64.S
+++ b/purgatory/arch/x86_64/setup-x86_64.S
@@ -42,10 +42,10 @@ purgatory_start:
/* In 64bit mode the code segment is meaningless */
movq 0(%rsp), %rax
- movq %rax, jump_back_entry
+ movq %rax, jump_back_entry(%rip)
/* Setup a stack */
- movq $lstack_end, %rsp
+ leaq lstack_end(%rip), %rsp
/* Call the C code */
call purgatory
